FeedbackGraph MCP server

Your bug evidence, inside your AI agent

Pull any report with full debugging evidence — or the top revenue-backed opportunities — into Claude Code, Cursor, VS Code, Windsurf or a Slack agent. Without leaving your tool.

Read the docsGet a key

Read-first · tenant-isolated · PII-redacted · audit-logged

AI agent · FeedbackGraph MCP
Pull report rep_8f2 with full evidence
fg_get_reportreports:read
Checkout 500 on card paymentbugHighAI 94%
consolepayment failed · [redacted-email]
networkPOST /api/pay → 500 · 12ms
jsErrorcharge declined · [redacted-card]
replay▶ app.acme.com/r/8f2
Top revenue-backed opportunities?
fg_list_opportunitiesintel:read
Top opportunities by revenue
1SSO / SAML$1.8Mscore 74
2Bulk CSV export$640Kscore 61
3Audit-log retention$410Kscore 52

Pull a report with full evidence

The AI verdict beside the raw proof — console logs, the failing network call, JS errors and a replay link — so your agent debugs with everything attached.

Ask for revenue-ranked work

“What should we build next, and what is it worth?” — answered with ARR and named accounts, not opinions, right where a PM is already working.

Secure by default

One tenant per token, enforced by row-level security. Scoped, revocable keys. Evidence redacted on the way out. Every call audited.

How it works

How does the FeedbackGraph MCP server work?

Add the server to your agent, paste a token, and your reports and Evidence Graph become tools the agent can call — agent → MCP → your tenant’s data, under RLS.

.cursor/mcp.json
{
  "mcpServers": {
    "feedbackgraph": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://feedbackgraph.com/api/mcp",
        "--header", "Authorization: Bearer sk_live_…"
      ]
    }
  }
}
AI agent
Pull report rep_8f2 with full evidence
fg_get_reportreports:read
Checkout 500 on card payment
POST /api/pay → 500 · AI confidence 94%
console: payment failed · [redacted-email]
▶ replay app.acme.com/r/8f2
The toolset

What tools does the MCP server expose?

Read-first by design. Every tool declares the scope it needs; your token decides what it can do.

fg_list_reportsreports:read

Filter reports by status, severity, type or project — paginated.

fg_get_reportreports:read

One report with full evidence: console, network, JS errors, replay, AI verdict.

fg_searchreports:read

Semantic search across reports and signals (pgvector cosine).

fg_list_opportunitiesintel:read

Revenue-ranked opportunities: score, ARR, accounts, stage.

fg_get_opportunityintel:read

Opportunity dossier: ARR, affected accounts, backing signals, next step.

fg_route_reportreports:write

Flag-gated: route a report to its tracker and get the URL back. Idempotent.

Resources:report://{id}opportunity://{id}— stable handles an agent can attach as context.
Works with
Claude CodeCursorVS CodeWindsurfSlack agents

Any client that speaks the Model Context Protocol — over stdio for local editors or streamable HTTP for remote and Slack agents.

Want to see the data before you wire it up?

Click through the same report and revenue-ranked opportunities in our 60-second interactive demo — no login, no backend.

Try the live demo
Security

Is the MCP server secure?

The MCP server reuses the exact authorized read paths your dashboard uses. It is tenant-facing only — there is no super-admin or cross-tenant surface, and no owner-database path an agent can reach.

  • One tenant per token, enforced at the database with row-level security
  • Scoped, revocable API tokens — reports:read, intel:read, reports:write
  • Evidence PII-redacted on output; media links stay expiring & signed
  • Per-token rate limits and a full audit log of every tool call
  • Read-first: the single write action is off by default behind a flag
  • Fail-closed: an unknown scope or cross-tenant id is denied, never partial data
Quickstart

How do I connect my AI agent?

Mint a key, drop this into your agent, and start asking. Full reference — tools, HTTP, scopes — in the docs.

.mcp.json (Claude Code)
{
  "mcpServers": {
    "feedbackgraph": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://feedbackgraph.com/api/mcp",
        "--header", "Authorization: Bearer sk_live_…"
      ]
    }
  }
}
Full developer docs

Questions

Is it secure? What can an agent actually see?+

A token resolves to exactly one tenant and every tool runs under Postgres row-level security — there is no cross-tenant or admin path. Evidence is PII-redacted on the way out, tokens are scoped and revocable, calls are rate-limited and audit-logged, and the server is read-first (the one write action is off by default behind a flag).

Which agents and editors work with it?+

Anything that speaks the Model Context Protocol — Claude Code, Cursor, VS Code, Windsurf, and remote/Slack agents. Local editors connect over stdio (via the mcp-remote bridge); remote agents connect to the streamable-HTTP endpoint with a Bearer token.

How do I get a token?+

Create an API key in Settings → API keys with the scopes you want (reports:read for reports and search, intel:read for opportunities, reports:write to enable routing). The key is shown once — drop it into your agent config.

Does it cost extra?+

No — the MCP server is included. It reads the same data your dashboard already shows, through the same authorized paths.

Bring your evidence into your agent

It’s already built and live. Get a key and your reports and revenue intelligence become callable in minutes.

Start freeRead the docs