# Fin integration
Source: https://help.craftcx.com/ai-agents/fin
Configure CraftCX scoring for Fin conversations
Configure Fin in CraftCX to attribute Fin messages and score eligible conversations.
## What CraftCX records
* Fin participation in a conversation
* AXIS results for eligible conversations
* Score explanations and conversation evidence
## Configure Fin
Connect your helpdesk and select Fin as the AI agent in CraftCX. CraftCX scores a conversation after it receives the required conversation events.
[Read about Fin](https://fin.ai?utm_source=craftcx\&utm_campaign=docs).
# Front Autopilot
Source: https://help.craftcx.com/ai-agents/front-autopilot
Identify and score Front Autopilot conversations
CraftCX identifies Front Autopilot messages from Front author metadata. It records them as AI messages, so you do not need to select a teammate.
CraftCX identifies messages from the `front_ai_autopilot` author. It tracks Autopilot Resolve as a separate AI agent.
## What CraftCX records
* Autopilot messages from supported Front authors
* AXIS results for resolution accuracy, interaction effort, and handoff smoothness
* AI participation in ticket score breakdowns
* Separate results for Autopilot Resolve, Front Chatbot, and human handoffs
## Configure Front Autopilot
Connect Front, enable Front Autopilot in CraftCX, and configure the [Front webhook rule](/integrations/front). CraftCX identifies supported Autopilot messages after Front sends the conversation.
[Read about Front AI](https://front.com/product/ai?utm_source=craftcx\&utm_campaign=docs).
# Front Autopilot Resolve
Source: https://help.craftcx.com/ai-agents/front-autopilot-resolve
Identify and score Front Autopilot Resolve conversations
CraftCX identifies Front Autopilot Resolve messages from Front author metadata. It records them as AI messages, so you do not need to select a teammate.
CraftCX identifies messages from the `front_companion` author. It tracks Front Autopilot as a separate AI agent.
## What CraftCX records
* Resolve messages from supported Front authors
* AXIS results for resolution accuracy, interaction effort, and handoff smoothness
* AI participation in ticket score breakdowns
* Separate results for Front Autopilot, Front Chatbot, and human handoffs
## Configure Front Autopilot Resolve
Connect Front, enable Front Autopilot Resolve in CraftCX, and configure the [Front webhook rule](/integrations/front). CraftCX identifies supported Resolve messages after Front sends the conversation.
[Read about Front AI](https://front.com/product/ai?utm_source=craftcx\&utm_campaign=docs).
# Front Chatbot integration
Source: https://help.craftcx.com/ai-agents/front-chatbot
Configure CraftCX scoring for Front Chatbot conversations
Configure Front Chatbot in CraftCX to attribute AI messages and score eligible conversations.
CraftCX identifies supported Front Chatbot messages after Front sends the conversation.
## What CraftCX records
* Front Chatbot participation in a conversation
* AXIS results for eligible conversations
* Score explanations and conversation evidence
## Configure Front Chatbot
Connect Front, select Front Chatbot in CraftCX, and configure the [Front webhook rule](/integrations/front).
[Read about Front AI](https://front.com/product/ai?utm_source=craftcx\&utm_campaign=docs).
# Gorgias AI integration
Source: https://help.craftcx.com/ai-agents/gorgias-ai
Configure CraftCX scoring for Gorgias AI conversations
Configure Gorgias AI in CraftCX to attribute AI messages and score eligible conversations.
## What CraftCX records
* Gorgias AI participation in a conversation
* AXIS results for eligible conversations
* Score explanations and conversation evidence
## Configure Gorgias AI
Connect Gorgias, then select Gorgias AI in CraftCX. CraftCX scores a conversation after Gorgias sends the required events.
[Read about Gorgias AI](https://gorgias.com/ai-agent?utm_source=craftcx\&utm_campaign=docs).
# AI agents
Source: https://help.craftcx.com/ai-agents/overview
Configure AI agents for your helpdesk automation
Configure an AI agent so CraftCX can identify its messages in your support conversations.
CraftCX attributes AI participation and scores eligible conversations.
Score Front Autopilot conversations.
Score Front Autopilot Resolve conversations.
Score Front Chatbot conversations.
Score Fin conversations.
Score Gorgias AI conversations.
Score Parahelp conversations.
## In development
These integrations are not yet available:
* Ada
* Forethought
* Zendesk AI
[Contact CraftCX support](mailto:support@craftcx.com) to request another AI agent.
# Parahelp integration
Source: https://help.craftcx.com/ai-agents/parahelp
Configure CraftCX scoring for Parahelp conversations
Configure Parahelp in CraftCX to attribute AI messages and score eligible conversations.
## What CraftCX records
* Parahelp participation in a conversation
* AXIS results for eligible conversations
* Score explanations and conversation evidence
## Configure Parahelp
Connect your helpdesk and select Parahelp as the AI agent in CraftCX. CraftCX scores a conversation after it receives the required events.
[Read about Parahelp](https://parahelp.com?utm_source=craftcx\&utm_campaign=docs).
# CraftCX MCP server
Source: https://help.craftcx.com/ai/mcp
Connect Cursor, Claude, Codex, and other MCP clients to CraftCX support intelligence and QA data
The CraftCX MCP server connects compatible AI assistants and coding agents to your CraftCX data through the [Model Context Protocol](https://modelcontextprotocol.io/). Connect it to search conversations, findings, customer signals, and support performance from your AI client.
CraftCX uses OAuth, limits each connection to one organization, and allows read-only access.
## Connect a client
1. Add the CraftCX remote MCP server to your AI client:
```text theme={null}
https://mcp.craftcx.com/mcp
```
2. Run your client’s MCP authentication command and sign in to CraftCX.
Your client can use CraftCX tools after you approve access to an organization.
## Endpoint
| Transport | URL |
| --------------- | ----------------------------- |
| Streamable HTTP | `https://mcp.craftcx.com/mcp` |
Read the protected resource metadata at:
```text theme={null}
https://mcp.craftcx.com/.well-known/oauth-protected-resource/mcp
```
## Authentication
CraftCX MCP uses OAuth. When you connect a client, it opens a browser flow so you can sign in to CraftCX and approve access for an organization.
You do not create an API token for MCP. Authorization uses your CraftCX account and the scopes you approve during consent.
## Permissions
CraftCX grants access through these scopes:
| Scope | What it unlocks |
| ------------------------- | ------------------------------------------------------ |
| `conversations:read` | Support conversations, messages, scores, and topics |
| `findings:read` | QA findings and operational incidents |
| `signals:read` | Customer signals and signal groups |
| `supportPerformance:read` | Aggregate AXIS scores, resolution outcomes, and trends |
Tools only appear when the session includes the matching scope. The MCP server is currently read-only.
## Available tools
| Tool | Description | Example use |
| ----------------------------- | ---------------------------------------------------------------------------------------------- | --------------------------------------------------------- |
| `whoami` | Return the authenticated user, organization, and approved scopes | Confirm which organization and scopes the session can use |
| `list-signals` | List recent customer signals | See what surfaced most recently |
| `search-signals` | Search persistent customer signals by text, type, and date | Find refund or shipping signals in a date range |
| `get-signal-group` | Retrieve a signal group with trends and representative evidence | Inspect a recurring customer issue |
| `search-signal-groups` | Search signal groups by text and type, or list recent groups | Find groups related to a product area |
| `search-conversations` | Search conversations by text, date, AXIS score, topic, and resolution status | Find low-scoring refund conversations |
| `get-conversation` | Retrieve a full conversation with messages, evaluations, scores, topics, findings, and signals | Dig into one conversation with evidence |
| `analyze-support-performance` | Analyze aggregate AXIS scores, resolution outcomes, and trends | Summarize quality over the last 30 days |
| `search-findings` | Search QA findings and incidents by text, priority, source, or status | List open high-priority findings |
| `get-finding` | Retrieve a finding or incident with evidence and related conversations | Review one finding end to end |
| `analyze-findings` | Summarize emerging and unresolved findings and incidents | Ask what needs attention this week |
Learn more about the product concepts these tools return:
* [AXIS scoring](/features/axis-score)
* [Audit](/features/audit)
* [Support Observability](/support-observability/overview)
* [Policy monitoring](/support-observability/policy-monitoring)
* [Incidents](/support-observability/incidents)
## Configure a client
Most agents are smart enough for you to simply ask them;
> *Add the CraftCX MCP server. Use the remote URL `https://mcp.craftcx.com/mcp`.*
### Cursor
1. Open **Cursor Settings** → **MCP** → **Add new global MCP server**
2. Paste:
```json theme={null}
{
"mcpServers": {
"craftcx": {
"url": "https://mcp.craftcx.com/mcp"
}
}
}
```
3. Save and restart Cursor if needed. Complete the OAuth flow the first time you use a CraftCX tool.
To share the config with your team, add the same block to `.cursor/mcp.json` in the project root.
### Claude Code
```bash theme={null}
claude mcp add --transport http craftcx https://mcp.craftcx.com/mcp
```
Then run `/mcp` in Claude Code and complete the OAuth flow.
### Codex
Add the server with the CLI:
```bash theme={null}
codex mcp add craftcx --url=https://mcp.craftcx.com/mcp
```
Or add the server to your `~/.codex/config.toml` file:
```toml theme={null}
[mcp_servers.craftcx]
url = "https://mcp.craftcx.com/mcp"
```
Then authenticate:
```bash theme={null}
codex mcp login craftcx
```
Complete the OAuth flow when prompted.
### OpenCode
CraftCX publishes an OpenCode remote config:
```text theme={null}
https://craftcx.com/.well-known/opencode
```
That endpoint advertises the CraftCX MCP server. OpenCode loads it when you authenticate against a provider that supports [well-known remote config](https://opencode.ai/docs/config/).
You can also add the server by hand in `opencode.json` or `opencode.jsonc`:
```json theme={null}
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"craftcx": {
"type": "remote",
"url": "https://mcp.craftcx.com/mcp",
"oauth": {}
}
}
}
```
Then authenticate:
```bash theme={null}
opencode mcp auth craftcx
```
Complete the OAuth flow when prompted. See the [OpenCode MCP docs](https://opencode.ai/docs/mcp-servers/) for more options.
### VS Code (GitHub Copilot)
Create `.vscode/mcp.json` in your workspace:
```json theme={null}
{
"servers": {
"craftcx": {
"type": "http",
"url": "https://mcp.craftcx.com/mcp"
}
}
}
```
Open the Command Palette, run **MCP: List Servers**, start the CraftCX server, and complete OAuth when prompted.
### Claude Desktop
1. Open **Settings** → **Connectors**
2. Add a connector with URL `https://mcp.craftcx.com/mcp`
3. Complete the OAuth flow
### Other clients
If your tool supports remote MCP over Streamable HTTP, use this URL:
```text theme={null}
https://mcp.craftcx.com/mcp
```
If the client only supports local stdio servers, bridge with [`mcp-remote`](https://www.npmjs.com/package/mcp-remote):
```json theme={null}
{
"mcpServers": {
"craftcx": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://mcp.craftcx.com/mcp"]
}
}
}
```
## Example prompts
After you connect a client, try these prompts:
* “What are the open high-priority findings in CraftCX?”
* “Search CraftCX for conversations about refunds with low AXIS scores this month.”
* “Summarize support performance for the last 30 days.”
* “Show recent customer signal groups related to shipping delays.”
* “Open the latest refund conversation and explain its findings and signals.”
## Troubleshooting
* Complete the OAuth consent flow and select the correct organization.
* If tools are missing, confirm the session approved the matching `*:read` scopes.
* Disconnect and reconnect in your client’s MCP settings if authentication fails.
* Confirm you can sign in to [app.craftcx.com](https://app.craftcx.com) with access to that organization.
# Create AI agent
Source: https://help.craftcx.com/api-reference/agents/create-ai-agent
/api-reference/openapi.json post /v1/agents
Create a custom AI agent that can be referenced by /assess.
# Authentication
Source: https://help.craftcx.com/api-reference/authentication
Authenticate requests to the CraftCX API
Authenticate every API request with a bearer token. Create and manage tokens in [Settings → API Tokens](https://app.craftcx.com/settings/api-tokens).
```bash theme={null}
curl https://api.craftcx.com/v1/agents \
-H "Authorization: Bearer $CRAFTCX_API_TOKEN"
```
# Create evaluation export
Source: https://help.craftcx.com/api-reference/export-data/create-evaluation-export
/api-reference/openapi.json post /v1/evaluations/exports
Queue an evaluation export job
# Download evaluation export
Source: https://help.craftcx.com/api-reference/export-data/download-evaluation-export
/api-reference/openapi.json get /v1/evaluations/exports/{id}/download
Redirect to the signed URL for a completed export. If Accept: application/json is sent, returns JSON with the download_url instead.
# Get evaluation export
Source: https://help.craftcx.com/api-reference/export-data/get-evaluation-export
/api-reference/openapi.json get /v1/evaluations/exports/{exportId}
Retrieve the status of an evaluation export job
# List evaluation exports
Source: https://help.craftcx.com/api-reference/export-data/list-evaluation-exports
/api-reference/openapi.json get /v1/evaluations/exports
Retrieve evaluation exports for the organization with pagination
# API reference
Source: https://help.craftcx.com/api-reference/introduction
Reference documentation for the CraftCX API
Use the CraftCX API to send data to CraftCX and retrieve evaluations.
## OpenAPI specification
The [OpenAPI specification](/api-reference/openapi.json) lists endpoints, request fields, and response fields.
# Score conversation
Source: https://help.craftcx.com/api-reference/score-conversation/score-conversation
/api-reference/openapi.json post /v1/assess
Send conversational message data from any source for AXIS scoring and analysis
# Get ticket
Source: https://help.craftcx.com/api-reference/tickets/get-ticket
/api-reference/openapi.json get /v1/tickets/{id}
Get a ticket by ID
# FAQ
Source: https://help.craftcx.com/faqs
Answers to common questions about AXIS conversation scoring
Find answers about AXIS scoring, helpdesk connections, exports, and support.
### How is the AXIS score calculated?
CraftCX rates each conversation from 1 to 5 for Resolution Accuracy, Interaction Effort, and Handoff Smoothness. The AXIS score is their average. Refer to the [AXIS scoring guide](/features/axis-score#how-axis-scoring-works) for more details.
### Which helpdesks are supported?
CraftCX supports any platform that can export transcripts. Front, HubSpot, Intercom, and Gorgias have direct integrations. The [API](/api-reference/introduction) supports custom sources.
[Contact CraftCX support](mailto:support@craftcx.com) to request another helpdesk.
### Can I export my results?
Yes. Use the dashboard or [API](/api-reference/export-data/create-evaluation-export) to export scores and explanations.
### Where do I get support?
Use the in-app support page or email [CraftCX support](mailto:support@craftcx.com).
# Article usage analytics
Source: https://help.craftcx.com/features/article-usage
Review the help center articles your AI shares with customers
Article Usage Analytics shows which help center articles your AI agent shares in customer conversations.
Use it to invest in articles that help customers and improve the content that creates extra work.
## Configure tracked domains
1. Open [Organization Settings](https://app.craftcx.com/settings/organization).
2. Add each help center domain that you want CraftCX to track.
3. Let your AI share articles from those domains in customer conversations.
CraftCX tracks a URL when your AI shares it in an eligible conversation.
The table covers the last 30 days of tracked article use.
## Read the table
The table lists each article URL, the number of conversations that included it, and its AXIS results.
Use the total score to compare the article's Resolution Accuracy, Interaction Effort, and Handoff Smoothness. The total score is their average.
## Act on the results
Review articles with high use and low scores first. Read the linked conversations to find missing information, outdated steps, or unclear wording.
Keep articles with high use and high scores as reference material when you write related content. Review topics that create escalations without article use for content gaps.
## FAQ
Add your help center domains, then confirm that your AI has shared URLs from those domains in eligible conversations.
CraftCX counts a URL when the AI shares it in a customer conversation and the URL matches a
configured domain.
Article Usage Analytics requires a Pro or Enterprise subscription. You can manage your plan in [Billing Settings](https://app.craftcx.com/settings/billing).
# Audit
Source: https://help.craftcx.com/features/audit
Review resolution status, AI participation, and AXIS scores
Audit gives you an independent record of AI participation and resolution status for each evaluated conversation.
Use it to verify resolution claims, investigate billing questions, and see where AI support needs attention.
## What Audit records
Audit classifies a conversation as resolved, partial, or not resolved. Use this result to review a vendor's resolution charge or a support workflow.
Audit classifies a conversation as AI only, AI assisted, or no AI. This shows how the AI
contributed to the outcome.
Audit includes the [AXIS score](/features/axis-score) and its Resolution Accuracy, Interaction Effort, and Handoff Smoothness scores.
## Review the audit dashboard
Filter conversations by date, resolution status, or AI participation. Open a conversation to read its transcript, score breakdown, and audit explanation.
Use filters before you export data. This limits an export to the conversations you need to review.
## Export audit data
1. Select a date range and any filters.
2. Select **Export**.
3. Download the CSV file when CraftCX finishes the export.
Each export can cover up to 367 days.
The CSV can include these fields:
* `ticket_id` and `helpdesk_id`
* `subject` and `channel`
* `resolution_status` and `ai_involvement`
* `axis_score`, `ra`, `ie`, and `hs`
* `resolution_note` and `audited_at`
Use the [CraftCX API](/api-reference/introduction) to create and manage evaluation exports from your own system.
## Review billing or performance
For billing, filter to the billed period and review the resolution status with the transcript. For performance, filter by AI participation and compare AXIS scores across conversation types or time ranges.
## FAQ
Audit includes conversations that CraftCX has evaluated.
Resolution status describes the customer outcome. AI participation describes the AI's role in the
conversation.
Export the relevant conversations and review their audit results with the transcript. Your vendor sets its own billing policy.
# AXIS conversation scoring
Source: https://help.craftcx.com/features/axis-score
Measure the quality of AI-powered support conversations
The AI Experience Impact Score (AXIS) measures the quality of each eligible AI support conversation. It turns individual conversations into a consistent measure of customer experience, so you can prove what works and target the gaps that matter.
## What AXIS measures
AXIS combines three scores into one total score.
Did the AI give the correct answer or route the customer correctly?
Review low scores when your knowledge base lacks information, contains outdated information, or sends customers to the wrong path.
How much work did the customer need to do to resolve the issue?
Review low scores when customers repeat information, receive unclear answers, or take too many steps.
Did a human handoff preserve the context the customer and agent need?
Review low scores when the AI transfers too late, misses key context, or sends the customer to the wrong team.
The total AXIS score is the average of these three scores.
## Read the results
Use the score explanation with the conversation transcript. A low score identifies a conversation to review. Repeated explanations can identify a knowledge, workflow, or AI configuration problem.
Compare scores by conversation type, time range, or AI configuration. Use the same comparison when you test a change.
## Improve a low score
1. Open a low-scoring conversation.
2. Read the score explanation and transcript.
3. Identify the lowest-scoring AXIS dimension.
4. Update the relevant knowledge, AI instruction, workflow, or handoff rule.
5. Compare later scores for the same type of conversation.
## FAQ
CraftCX creates a score after your helpdesk marks an eligible conversation as complete. The timing depends on the helpdesk events CraftCX receives.
CraftCX cannot score a conversation without enough usable conversation data. Contact [CraftCX
support](mailto:support@craftcx.com) if an expected score is missing.
AXIS uses standard criteria so that you can compare scores across conversations and over time.
AXIS measures the AI part of the customer experience. It does not score individual human agents.
Group related low scores into recurring failure patterns.
Review resolution status and AI participation with AXIS results.
# Failure clusters
Source: https://help.craftcx.com/features/failure-clusters
Group related AI support failures so you can fix their shared cause
Failure Clusters group related low-scoring AI conversations.
Use a cluster to find the shared cause of a recurring problem instead of reviewing each conversation separately.
## What CraftCX groups
CraftCX groups conversations by their topic, failure type, score impact, and recurrence. A cluster includes representative conversations that show the shared problem.
Clusters can identify inaccurate knowledge, repeated clarification, unclear workflows, or weak handoffs.
## Prioritize a cluster
Start with clusters that contain more conversations or have a larger AXIS impact. Read the summary, then open representative conversations to confirm the pattern.
Review whether the failure affects Resolution Accuracy, Interaction Effort, or Handoff Smoothness. That dimension points to the change you need.
## Fix the shared problem
1. Open a high-impact cluster.
2. Read representative conversations and their score explanations.
3. Identify the shared cause.
4. Update the relevant knowledge, AI instruction, workflow, or handoff rule.
5. Watch later conversations to confirm that the cluster shrinks.
## FAQ
CraftCX creates a cluster only when related failures recur. You may have isolated failures or too little conversation data.
CraftCX uses consistent grouping criteria so that you can compare clusters over time.
Compare the cluster's later conversation count and AXIS scores with the results before your change.
# Topic detection
Source: https://help.craftcx.com/features/topic-detection
Group support conversations by the subjects customers discuss
Topic Detection groups conversations by the subjects customers discuss.
Use topic volume and AXIS results to see customer demand and prioritize improvements to AI support, documentation, or product workflows.
## What CraftCX identifies
CraftCX identifies product areas, customer goals, problem descriptions, and request types in conversation content. A conversation can include more than one topic.
Topics can include product setup, billing, integrations, troubleshooting, feature requests, and account questions.
## Read topic results
Each topic shows its conversation volume and average AXIS scores. Use volume to understand demand and scores to understand conversation quality.
Compare topics over time after a product change, documentation update, or AI configuration change.
## Improve a topic
1. Find a topic with high volume and low AXIS scores.
2. Read representative conversations for that topic.
3. Identify the missing knowledge, unclear process, or handoff problem.
4. Update the relevant content, AI configuration, or workflow.
5. Compare later volume and AXIS scores.
## FAQ
Yes. CraftCX identifies the primary subjects that appear in a conversation.
CraftCX creates topics from your conversation content and customer language.
CraftCX analyzes a conversation after it receives the conversation data needed for evaluation.
# Front integration
Source: https://help.craftcx.com/integrations/front
Connect Front and configure its webhook rule
Connect Front to send eligible conversations to CraftCX for scoring.
1. In CraftCX, select **Integrations → New Integration → Front**.
2. In Front, grant access and allow **Shared Resources**.
3. Return to CraftCX and select **Create Webhook**.
4. In Front, create a rule that uses **Send to webhook** with the generated URL.
5. Select **Backfill** to score recent tickets.
## Configure the Front rule
For **Front Chat** channels that use AI Answers, run the rule on **Inbound message is received (new conversation)**.
For another AI provider in Front, run the rule when you hand off to a human or close the conversation.
CraftCX ingests and scores new conversations after the rule runs.
## Configure your AI agent
Select the AI agent that handles conversations in your helpdesk.
1. Open **AI Agent Settings**.
2. Select your AI agent provider.
3. Complete any required fields and select **Save**.
# Gorgias integration
Source: https://help.craftcx.com/integrations/gorgias
Connect your Gorgias account to CraftCX
Connect Gorgias to send eligible conversations to CraftCX for scoring.
1. In CraftCX, select **Integrations → New Integration → Gorgias**.
2. Enter your subdomain and approve the Gorgias connection.
3. Return to CraftCX and confirm that the account appears in your integrations list.
4. Select **Backfill** to score recent tickets.
## Configure your AI agent
Select the AI agent that handles conversations in your helpdesk.
1. Open **AI Agent Settings**.
2. Select your AI agent provider.
3. Complete any required fields and select **Save**.
# HubSpot integration
Source: https://help.craftcx.com/integrations/hubspot
Connect your HubSpot account to CraftCX
Connect HubSpot to send eligible conversations to CraftCX for scoring.
1. In CraftCX, select **Integrations → New Integration → HubSpot**.
2. In HubSpot, approve the connection.
3. Return to CraftCX and confirm that the account appears in your integrations list.
## Configure your AI agent
Select the AI agent that handles conversations in your helpdesk.
1. Open **AI Agent Settings**.
2. Select your AI agent provider.
3. Complete any required fields and select **Save**.
# Intercom integration
Source: https://help.craftcx.com/integrations/intercom
Connect your Intercom workspace to CraftCX
Connect Intercom to send eligible conversations to CraftCX for scoring.
1. In CraftCX, select **Integrations → New Integration → Intercom**.
2. In Intercom, select a workspace and grant the requested permissions.
3. Return to CraftCX and confirm that the workspace appears in your integrations list.
4. In Intercom, configure a webhook to send conversation events to your CraftCX endpoint.
CraftCX ingests and scores conversations after Intercom sends events.
## Configure your AI agent
Select the AI agent that handles conversations in your helpdesk.
1. Open **AI Agent Settings**.
2. Select your AI agent provider.
3. Complete any required fields and select **Save**.
# Integrations
Source: https://help.craftcx.com/integrations/overview
Connect your helpdesk to CraftCX
Connect Front, HubSpot, Intercom, or Gorgias to send conversations to CraftCX.
Use the [CraftCX API](/api-reference/introduction) to upload transcripts from another source.
Select an integration for setup steps.
Connect your Front account.
Connect your Gorgias account.
Connect your HubSpot account.
Connect your Intercom account.
## AI assistants
Connect Cursor, Claude, Codex, and other MCP clients to read CraftCX support intelligence through the [CraftCX MCP server](/ai/mcp).
# Overview
Source: https://help.craftcx.com/introduction
Understand and improve the quality of AI-powered customer support
## What CraftCX does
CraftCX evaluates the conversations your AI agents handle. It shows what happened, how customers experienced it, and what your team can improve.
Use CraftCX alongside metrics such as deflection, containment, and response time. Review conversation evidence and patterns across your support operation.
## Features
Evaluate resolution accuracy, customer effort, and handoff quality with the AXIS score.
See the reasoning, conversation evidence, and failure themes behind each assessment.
Compare performance over time to find recurring problems and quality regressions.
Ingest conversations from your helpdesk or send transcripts through the CraftCX API.
Connect Front, HubSpot, Intercom, or Gorgias. You can also send transcripts through the [CraftCX API](/api-reference/introduction) or connect AI assistants through the [CraftCX MCP server](/ai/mcp).
## Support Intelligence
As AI agents resolve more conversations, important customer insight can stay hidden in support transcripts. CraftCX preserves that insight by detecting signals such as bug reports, feature requests, documentation gaps, usability problems, and billing issues.
Related signals are grouped into recurring customer problems and needs, so your team can:
* Spot new and growing issues
* Understand which problems demand attention
* Review the conversations behind each signal
* Turn support activity into evidence for product and operational decisions
[Learn more about Signals](/support-intelligence/signals), or [explore Signals in CraftCX](https://app.craftcx.com/signals).
## Who uses CraftCX
CraftCX is built for teams responsible for the quality and performance of AI-powered support.
Understand whether AI is delivering the experience your customers expect.
Find failure patterns, monitor changes, and decide what to improve next.
Evaluate agent behavior with independent evidence and track quality over time.
Follow the [quickstart](/quickstart) to connect your first source.
# Invite your team
Source: https://help.craftcx.com/invite-team
Add members to your CraftCX workspace
Administrators manage team members from [Settings → Team](https://app.craftcx.com/settings/team).
1. Open the settings menu and select **Team**.
2. Enter the email of the teammate you want to invite.
3. Select their role and send the invitation.
CraftCX sends an email with instructions to join the workspace.
# Quickstart
Source: https://help.craftcx.com/quickstart
Connect a helpdesk and review your first AXIS scores
Connect a helpdesk, then review the scores CraftCX creates for AI-handled conversations.
## Connect your helpdesk
1. Create an account in [CraftCX](https://app.craftcx.com/sign-up).
2. Open **Integrations** and select **New Integration**.
3. Select your helpdesk and complete its connection flow.
CraftCX scores eligible conversations after your AI agent handles them. See [integrations](/integrations/overview) for provider-specific setup.
## Next steps
Share results so everyone can track improvements.
Manage integrations and API access.
Send transcripts from a custom source.
Learn more about how scoring works.
# Using settings
Source: https://help.craftcx.com/settings-page
Manage your CraftCX workspace and connected services
Use **Settings** to manage your workspace and its connected services.
You can:
* Connecting or disabling helpdesk integrations.
* Generating API tokens for programmatic access.
* Updating billing information and notification preferences.
CraftCX saves your changes automatically. Each setting includes help text when needed.
# Signals
Source: https://help.craftcx.com/support-intelligence/signals
Turn customer conversations into grouped evidence about bugs, requests, friction, and emerging operational issues.
Signals help you see what customers are experiencing across the conversations handled by your support operation. CraftCX detects meaningful observations in individual conversations, then groups related observations into recurring customer problems and needs.
This preserves the insight that support teams often develop by reading tickets every day: a new bug appearing after a release, repeated requests for the same capability, a confusing workflow, or a gap in your documentation.
[Open Signals in CraftCX](https://app.craftcx.com/signals).
## What CraftCX detects
Each signal has one type:
| Type | What it captures |
| --------------------- | -------------------------------------------------------------------- |
| **Bug report** | A customer describing behavior that appears broken or incorrect |
| **Feature request** | A request for a new capability or an improvement to an existing one |
| **Documentation gap** | Missing, unclear, or hard-to-find guidance |
| **Usability issue** | Friction or confusion caused by a workflow or interface |
| **Billing issue** | Problems with charges, plans, invoices, or payment workflows |
| **Incident signal** | Activity that may point to a broader service or operational incident |
A single conversation can produce more than one signal. For example, a customer asking how to export data might reveal both a feature request for self-service exports and a documentation gap around the current process. CraftCX keeps those observations separate so each underlying need remains clear.
## Signal groups
The **Signals** tab groups related observations into persistent themes. Each group represents one specific customer problem or desired outcome and includes:
* A title and summary of the recurring theme
* Its signal type, lifecycle, and trend
* Conversation volume for the last 30 days and overall
* Change compared with the previous 30 days
* When the theme was first and most recently seen
By default, groups are ranked using lifecycle, growth, recent activity, and conversation volume. You can also sort by the most recently seen groups or the groups with the most conversations.
### Lifecycle
Lifecycle shows how established a signal group is:
| Lifecycle | Meaning |
| --------------- | ----------------------------------------------------------- |
| **New** | A recently identified recurring theme |
| **Recurring** | A theme that continues to appear in customer conversations |
| **Established** | A sustained theme with an established history |
| **Inactive** | A theme that is no longer appearing in recent conversations |
Growing groups also receive a **Growing** trend label, making increases in customer activity easier to spot.
### Filtering groups
Use the controls at the top of the Signals page to filter groups by signal type or lifecycle. Filters can help you answer focused questions, such as:
* Which bug reports are new?
* Which feature requests keep recurring?
* Are any documentation gaps growing?
* Which established issues have become inactive?
## Reviewing source evidence
Open a signal group to see the evidence behind it. The detail page includes the group's summary, activity totals, first-seen and last-seen dates, and every supporting signal currently assigned to the group.
Each supporting signal links to its source conversation, so you can review the customer's language and the surrounding context before deciding what to do next.
Use source conversations to validate the scope of a problem. A group summarizes the shared theme,
while the underlying tickets show the different ways customers describe and experience it.
## The Signals Feed
The **Feed** tab shows individual signals as CraftCX detects them, before or alongside grouping. Each entry includes its type, summary, detection time, and a link to the source ticket.
Use the feed when you want to review the latest customer observations. Use grouped Signals when you want to understand repeated themes, changes in volume, and which problems may deserve broader attention.
## Using Signals in your workflow
Signals are designed to turn support activity into evidence for support, product, and operational decisions. A practical review loop is:
Review the highest-ranked themes for new, growing, or high-volume customer problems.
Filter by type or lifecycle when you are investigating a specific area, such as recent bugs or
recurring feature requests.
Open a group and read its supporting conversations to confirm the problem and understand its
impact.
Use the group's summary and source conversations as evidence when prioritizing product,
documentation, or operational work.
Review grouped customer problems and needs in CraftCX.
Learn how Signals fit into the CraftCX product.
# Incidents
Source: https://help.craftcx.com/support-observability/incidents
Track severe or recurring risks in AI-powered support
Incidents turn serious or recurring AI support risks into an owned queue.
CraftCX creates an incident for a severe finding or a problem that recurs across conversations, so the right team member can investigate and resolve it.
## What creates an incident
CraftCX can create an incident for:
* A recurring policy deviation
* A high-severity finding
An incident represents the shared problem. The linked findings and conversations provide its evidence.
## Triage an incident
1. Open the incident and read its summary, linked rule, and related conversations.
2. Assign the incident to the team member who owns the fix.
3. Set a due date when the work has a deadline.
4. Set the status to **Acknowledged**, **Needs fix**, **Resolved**, or **False positive**.
CraftCX records status changes with the time and team member who made them.
## Use the incident queue
Use **Unassigned** to find incidents that need an owner. Use **Assigned**, **Assigned to me**, and **Resolved** to review work by status.
An incident can be on track, at risk, or overdue based on its due date.
## FAQ
No. CraftCX creates incidents from policy findings and quality signals.
A finding records a policy deviation in one conversation. An incident records a severe finding or
a problem that recurs across conversations.
CraftCX keeps the incident in the Resolved view for your records. A later occurrence can create a new incident.
Review the policy findings that can create incidents.
Send incident alerts to your team.
# Notifications
Source: https://help.craftcx.com/support-observability/notifications
Send CraftCX findings and incidents to Slack, email, or a webhook
Notification rules send CraftCX findings to the people and systems that need them. They keep your team informed without requiring manual dashboard checks.
## Create a notification rule
1. Open **Settings → Notifications → Alerts**.
2. Select **Create notification rule**.
3. Enter a descriptive name.
4. Select one or more events.
5. Select immediate delivery or a scheduled digest.
6. Add one or more destinations.
7. Save the rule.
You can send a rule to Slack, email, or a webhook. A webhook destination receives a POST request with the event data.
## Choose events and schedules
Choose **Incident detected** to send new policy findings or incidents. Choose **Low AXIS score** to send a score that falls below your threshold.
Immediate delivery sends the notification when the event occurs. A scheduled digest groups events and sends one message on the days and hour you select.
## Manage notification rules
Open **Settings → Notifications → Alerts** to edit, disable, or delete a rule. Disabling a rule pauses delivery and keeps its settings.
Use **Settings → Notifications → Preferences** for personal in-app notifications. These preferences do not change shared rules.
## Configure destinations
Connect Slack in **Settings → Integrations** before you select a Slack channel. Enter an email address for email delivery. Enter a URL that accepts POST requests for a webhook destination.
## FAQ
Yes. Add each Slack channel, email address, or webhook to the rule.
Yes. Disable the rule in the alerts list, then enable it when you want delivery to resume.
CraftCX uses your browser timezone when you create or edit the rule.
# Support observability
Source: https://help.craftcx.com/support-observability/overview
Find policy violations, quality changes, and recurring support risks
Support Observability makes AI support risks visible in the conversations that create them. It evaluates conversations against your policies and records evidence.
Use it to turn scattered warning signs into findings and owned incidents your team can resolve.
## How it works
1. Connect a helpdesk and add your policy sources.
2. Review the rules CraftCX extracts from your policies.
3. CraftCX evaluates eligible AI-handled conversations against active rules.
4. CraftCX creates findings with conversation evidence and policy citations.
5. CraftCX groups severe or recurring findings into incidents.
Use notifications to send important findings and incidents to Slack, email, or a webhook.
## Use the results
Review a finding to confirm the policy deviation and its evidence. Mark false positives, assign work, and set due dates when an issue needs action.
Review incidents to track systemic problems such as a policy violation that recurs across conversations or a sustained quality regression.
Evaluate AI behavior against your policies.
Track severe and recurring issues.
Send findings and incidents to your team.
# Policy monitoring
Source: https://help.craftcx.com/support-observability/policy-monitoring
Evaluate AI support conversations against your policies and procedures
Policy Monitoring evaluates AI support conversations against the policies and procedures you define.
It helps you catch policy risks with the evidence to act: the relevant conversation messages, policy citations, severity, and explanation.
## Add policy sources
1. Open **Policy Docs**.
2. Upload a PDF or Word document, or connect a Notion workspace.
3. Review the draft rules CraftCX extracts from the source.
4. Activate the rules you want CraftCX to evaluate.
Clear source language produces clearer rules. Review each rule before you activate it.
## Review findings
CraftCX evaluates eligible AI-handled conversations against active rules. When CraftCX finds a deviation, it creates a finding.
Each finding includes:
* A summary and severity
* Conversation messages that show the deviation
* Citations to the relevant policy source
* Reason codes and policy risk categories
* A confidence rationale
## Triage a finding
Use these statuses to record your decision.
| Status | Use it when |
| -------------- | --------------------------------------------------------- |
| Acknowledge | You reviewed the finding. |
| Needs fix | Your policy, AI configuration, or process needs a change. |
| False positive | CraftCX flagged the conversation incorrectly. |
| Resolved | You fixed the underlying issue. |
Assign a finding and set a due date when another team member owns the work.
## Work with related incidents
CraftCX promotes severe findings and recurring deviations into [incidents](/support-observability/incidents). Use an incident to track the underlying problem across several conversations.
## FAQ
Policy Monitoring evaluates eligible AI-handled conversations from connected helpdesks. It does not evaluate human-only conversations.
[AXIS](/features/axis-score) measures conversation quality. Policy Monitoring checks whether the
AI followed your defined rules. A conversation can score well and still violate a policy.
Mark the finding as a false positive and clarify the source policy. CraftCX evaluates the active rule against the conversation evidence.