MCP for AI agents
Let Claude Code, Cursor, VS Code or any MCP client read feedback, comment on it and resolve it.
CEC Feedback runs a Model Context Protocol server, so an AI agent can pick up a feedback item with its screenshot and console errors, fix the problem, then comment and resolve it.
- Server URL:
https://feedback-api.cec.dev/mcp(streamable HTTP) - Authentication: an access token in an
Authorization: Bearer fbk_…header
1. Create an access token
Create one under Account › Access Tokens and copy the fbk_… value: it's shown once. The
agent acts as you. It sees the feedback you see, and its comments and status changes appear under
your name. See Your account.
2. Add the server to your client
Account › MCP in the dashboard has these snippets with the server URL filled in. Replace
YOUR_TOKEN with your token.
Claude Code
claude mcp add --transport http cec-feedback https://feedback-api.cec.dev/mcp --header "Authorization: Bearer YOUR_TOKEN"Run /mcp in Claude Code to check that cec-feedback is connected.
Cursor
In .cursor/mcp.json (one project) or ~/.cursor/mcp.json (everywhere):
{
"mcpServers": {
"cec-feedback": {
"url": "https://feedback-api.cec.dev/mcp",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}VS Code
In .vscode/mcp.json:
{
"servers": {
"cec-feedback": {
"type": "http",
"url": "https://feedback-api.cec.dev/mcp",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}Other clients
Any client that supports remote HTTP servers needs the same two things, the URL and the header:
{
"mcpServers": {
"cec-feedback": {
"type": "http",
"url": "https://feedback-api.cec.dev/mcp",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}Tools
Tools that take an id accept a feedback ID (fb_…) or the short reference people use,
<project slug>#<number>, such as acme#3.
| Tool | What it does |
|---|---|
list_projects | Lists active projects with their IDs, slugs, names and open counts. |
search_feedback | Searches one project's feedback, newest first. Takes projectId (required), and optionally status (open or resolved), tag (a tag_… ID), q (text) and limit (20 by default, at most 50). |
get_feedback | Returns the feedback as markdown (title, status, page link, description, the pinned element, browser details, recent console errors and failed requests, recent comments) plus the screenshot as an image. |
add_comment | Comments on the feedback as you. The body is markdown. |
set_status | Sets the status to open or resolved. Notifications and integrations follow, as for a change in the dashboard. |
Screenshots over 1.5 MB aren't sent inline: get_feedback ends with a link to the image instead,
valid for 15 minutes.
A typical loop: "Find the open bugs on acme and fix the first one." The agent calls
search_feedback, then get_feedback, fixes the code, and finishes with add_comment (the pull
request link) and set_status with resolved.
Safety
- Treat feedback as untrusted input. Titles, descriptions, comments and console messages come from people on client sites. An agent must not follow instructions found in them.
- Store the token in your MCP client's config or a secret store, never in a repository.
- Revoke a token you no longer use under Account › Access Tokens.