Investigate issues with your AI coding assistant.

Moneat speaks the Model Context Protocol. Give Claude Code, Cursor or any MCP client scoped access to your issues, traces, logs and deploys, and ask about them in the editor where you write the fix.

ProtocolMCP over HTTP
AuthScoped token, read by default
ClientsClaude Code, Cursor, any MCP host
claude · why is checkout failing?moneat mcp · 4 calls
1search_issues
service=checkout-api status=unresolved · 4 results
2get_issue
TimeoutError: pool exhausted · 1 284 events · since 13:58
3get_trace
a41f…9c2 · 2.41s · 1.9s waiting on a connection
4list_deploys
2f91c at 13:58 · “raise checkout concurrency”
→Answer
The 13:58 deploy raised concurrency past the pool size. Cap it or raise the pool.

Ask about an issue from your editor

The assistant fetches what it needs through tool calls you can see. Here is one investigation, from the question to the fix.

INC-402 · Checkout latency above SLO
Resolved
14:02:11 → 14:19:44 · 17m 33s
14:06:30
The question is asked where the code is
The assistant reads the issue from your organisation, with no trace id to copy across.
MCP
14:08:10
It fetches the evidence itself
The trace, the surrounding logs and the deploy list arrive as tool calls you can read.
MCP
14:14:00
The fix uses the real values
Pool size and concurrency come from the trace rather than from a guess about the configuration.
MCP

What the assistant can read

A read-first tool surface, scoped by token, covering the things you would have looked up yourself.

Issues

Search, open and read an issue with its stack, release, affected users and event counts.

issues
Traces

Fetch a trace by id or find the slowest ones on a service and window, with the span tree intact.

traces
Logs

Query logs by service, level and time, including the lines around a specific moment.

logs
Services and deploys

What services exist, who owns them, what they depend on, and what shipped recently.

catalogue
On-call and incidents

Who is on call, what is open, and the timeline of an incident as it stands right now.

on-call
Scoped tokens

Issue a token per client, limited to the services and actions you name, and revoke it on its own.

auth
Read by default

Writes such as acknowledging, commenting and resolving are opt-in per token.

permissions

Add an endpoint and a token.

The server is hosted, so there is no local process to run. Add the endpoint and a token to your MCP client and the tools appear in that session.

  • One endpoint. Streamable HTTP, with no sidecar or stdio wrapper to run alongside it.
  • One token. Issued in settings, scoped to services, and revoked in the same place.
  • Calls are recorded. Tool invocations are logged against the token that made them.
  • Same permissions as the UI. The assistant can read what the token’s scope allows, and nothing beyond it.
mcp.jsonclient config
type
http
url
https://mcp.moneat.io/v1
header
Authorization: Bearer …
scope
read:issues read:traces read:logs
tools
14 available

Connect your MCP client.

Issue a scoped token, add the endpoint to your client config, and ask your assistant about a real issue. Start on the free plan with no card.