AI Integration (MCP)
Crosscheck integrates with AI coding assistants through the Model Context Protocol (MCP). This allows AI tools to access your Crosscheck workspace data, helping you debug issues, analyze checks, and understand team feedback without leaving your development environment.
What is MCP?
The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to external tools and data sources. Instead of copying and pasting information between Crosscheck and your AI assistant, MCP creates a direct connection so the AI can query your workspace data programmatically.
Supported Tools
Crosscheck exposes four read-only tools for workspace users. Results contain readable text and, when available, screenshot images.
| Tool | Inputs | What it returns |
|---|---|---|
| list_projects | None | Projects visible to you in the authorized workspace, with names and IDs. |
| list_checks | Optional projectId, type, limit (default 20) | Recent checks with IDs, capture types, dates, projects, and tags. |
| get_check | Required checkId; optional image, console, and network filters | Capture metadata, screenshot or replay summary, linked tasks, and available developer context. |
| get_comments | Required checkId | Comments with authors, dates, video timestamps, and attachments. |
For get_check and get_comments, use a check ID or the public token from a shared link ending in /public/checks/<token>. A full URL is not a tool argument; your assistant must extract the ID or token first.
Focus the debugging context
| get_check option | Default | Purpose |
|---|---|---|
| include_image | true | Include the available screenshot or thumbnail. |
| errors_only | false | Return console errors instead of all console logs. |
| network_filter | all | Use errors to keep requests with HTTP status 400 or higher. |
| include_network_detail | true | Include request and response headers and bodies. |
| max_entries | 100 | Maximum entries per developer-context collection. |
For smaller responses, lower max_entries or turn off include_network_detail. Console and network data must have been captured with the check to appear in the result.
Prompt and resource
Clients that support MCP prompts can run debug_check with a checkId to assemble capture metadata, the available screenshot, console errors, failed requests, and comments. Clients that support resources can read crosscheck://projects for the workspace project list.
Setup
Server URL: https://mcp.crosscheck.cloud/mcp. Choose a client that supports remote Streamable HTTP and OAuth with PKCE. Setup differs by client; the examples below are for Cursor and Claude Code.
Cursor
Add this to .cursor/mcp.json in your project, or ~/.cursor/mcp.json for your user. Merge it with any existing servers.
{
"mcpServers": {
"crosscheck": {
"url": "https://mcp.crosscheck.cloud/mcp"
}
}
}Follow Cursor's authentication prompt. See the Cursor MCP documentation for client settings.
Claude Code
Add the remote server from your terminal:
claude mcp add --transport http crosscheck https://mcp.crosscheck.cloud/mcp
Run /mcp inside Claude Code, select Crosscheck, and complete authentication. See the Claude Code MCP documentation for additional configuration options.
Authorize a workspace
Verify the connection
Example Use Cases
Here are some example prompts you can use with your AI assistant once Crosscheck is connected:
- "Show me the most recent bug reports in the Dashboard project."
- "List my recent checks, then inspect their console errors."
- "Analyze the failed network requests in this check: <paste a Crosscheck check link>."
- "Summarize the team comments on the latest screen recording."
- "What JavaScript errors were captured in the most recent instant replay?"
Usage Limits
| Limit | Current behavior |
|---|---|
| Free workspace | 1,000 accepted MCP POST requests per calendar month (UTC), shared across workspace clients and users. |
| Paid workspace | No monthly MCP request cap. |
| Short-term rate limit | 60 POST requests per minute per IP address, including paid workspaces. |
Initialization and tool discovery requests also count toward the free allowance when accepted by the server. Viewer access requires a paid workspace. Existing resource permissions and retention rules still apply.
Troubleshooting
| Symptom | What to check |
|---|---|
| 401: invalid or expired OAuth token | Reauthenticate through your client. Do not substitute a REST session token. |
| 403: workspace access denied | Confirm your membership is active. VIEWER_ACCESS_REQUIRES_PRO means viewer access is paused on a free workspace. |
| 429: MCP_MONTHLY_LIMIT_REACHED | The workspace has used its monthly allowance. Wait for the next UTC calendar month or upgrade. |
| 429: rate limit exceeded | Reduce request frequency and retry after the short-term limit resets. |
| 405 from GET or DELETE /mcp | Use a Streamable HTTP client. The server accepts POST and does not offer a persistent SSE stream. |
| Missing screenshot or developer context | Confirm that data was captured, processing has finished, and the check is within its retention window. |
Privacy and Security
The Crosscheck MCP integration is designed with security in mind:
- Workspace-scoped — Project and check lists use the workspace you authorize and your existing access permissions. Detail tools can also resolve a valid public check token from a shared link.
- Read-only — The MCP integration provides read-only access. Your AI assistant cannot modify, delete, or create checks or other data in Crosscheck.
- OAuth secured — All connections are authenticated through OAuth with your Crosscheck account. Removing the server from your AI client stops that client from using the connection; this does not itself revoke issued tokens.