Skip to content
Fixter Docs
↑↓ Select⏎ OpenEsc Close

Fixter MCP server

The Fixter MCP server gives your coding agent your production data: logs, traces, metrics, and the errors Fixter groups into issues. Ask what broke, set up alerts, or pick up an investigation without leaving your editor. It works with Claude Code, Cursor, and any MCP client.

What you can ask

Section titled “What you can ask”

Once connected, you ask in plain language and the agent picks the tools. Some real prompts and what they run on:

You typeWhat the agent uses
”why did checkout fail last night?”logs, spans, get_trace, and correlate to find the failing requests and follow them across services
”what new errors showed up since this morning’s deploy?”list_issues with status NEW, sorted by first seen
”what’s the root cause of this issue?”get_issue, which includes Fixter’s root-cause analysis when an investigation has run, then the traces and logs behind it
”which users hit this error since the deploy?”run_sql over the logs or spans behind the issue
”show the slowest endpoints today”aggregate_spans, p95 latency grouped by operation
”set up an alert when payment errors spike”describe_alerting, preview_alert_rule to test it against recent data, then save_alert_rule
”make a dashboard of checkout error rate and p95 latency”mint_dashboard, which returns a link you can share
”pick up investigation clever-nimble-otter-31”start_investigation, which hands your agent the investigation’s full transcript

Answers come from the telemetry you send. A question about users finds them only if your logs or spans carry a user id. Alerts from a rule reach you by email, and in Slack once Slack is connected.

What your agent gets

Section titled “What your agent gets”

The tools fall into groups. The MCP tools reference lists every one with its parameters.

  • Telemetry and querying: search logs and spans, fetch whole traces, query metrics, aggregate latency and error rates, and run read-only QuerySQL. Telemetry and querying tools.
  • Dashboards: turn a query into a dashboard with a stable link, and update it later. Dashboards tools.
  • Alerting: write, preview, enable or disable, and delete alert rules, and choose which email or Slack channels each rule’s alerts go to. Alerting tools.
  • Issues and noise: list and read the errors grouped from your telemetry and the routes and jobs they hit, then resolve or suppress them. Silence what should not page you with suppression rules, alert suppressions, and ignore rules. Issues and noise tools.
  • Investigations: list recent investigations, read one, or claim it and continue in your editor. fix_issue hands an issue to Fixter’s fix agent, which opens a pull request on your repository; this is in beta, a person at Fixter reviews each one, and Fixter never merges. These tools also set up notification channels and where each kind of notification goes. Investigations tools.
  • API tests and uptime: create probes against your HTTP endpoints and MCP servers, read their run history, check uptime, and choose where their alerts go. API & MCP testing. API tests and uptime tools.
  • Account: get the endpoint and a one-time key link for sending telemetry, and list the ingestion API keys issued so far. Account tools.

You can choose which of these tools your agent sees, as a per-user preference under User settings > Tools; see Managing MCP tool list.

Telemetry queries only read. Tools that change something (alert rules, dashboards, issue status, API tests) act immediately, and your agent acts with your permissions, or the key’s.

In Claude Code, the Claude Code plugin runs on top of this server: its diagnose and alert-setup skills drive these same tools.

Connect

Section titled “Connect”

Connect your editor

Section titled “Connect your editor”
claude mcp add fixter --transport http https://mcp.fixter.dev

Then run /mcp inside Claude Code. A browser window opens and you sign in with your Fixter account. The Fixter tools stay unavailable until you do.

/mcp

To share the setup with your team, commit a .mcp.json at the repo root instead:

{
  "mcpServers": {
    "fixter": {
      "type": "http",
      "url": "https://mcp.fixter.dev"
    }
  }
}

Connect a headless agent or CI (API key)

Section titled “Connect a headless agent or CI (API key)”

An agent with no browser, such as a CI job or a server-side automation, uses an API key instead. Create one at Settings > API keys (the value is shown once, so copy it then) and send it as a header. In Claude Code:

claude mcp add fixter --transport http https://mcp.fixter.dev \
  --header "Authorization: Bearer fixter_sk_<your-key>"

Other clients send the same Authorization: Bearer fixter_sk_<your-key> header on every request. Keys do not expire on their own, so use one only where nobody can sign in.

Verify it works

Section titled “Verify it works”

Ask your agent to list its Fixter tools. You should see the telemetry tools (logs, spans, get_trace, run_sql, and the rest). Then ask it something real:

Using Fixter, what services logged errors in the last hour?

If the tools don’t appear

Section titled “If the tools don’t appear”
  • Endpoint: check the URL is exactly https://mcp.fixter.dev. Setups that use the older https://mcp.fixter.dev/mcp keep working.
  • Sign-in pending: in Claude Code, adding the server does not sign you in. Run /mcp and finish signing in in the browser. In other clients, complete the browser sign-in if one is waiting.
  • Revoked key: for key-based setups, confirm the key is still active. Settings > API keys lists active keys by prefix.
  • Truncated results: long results are cut short on purpose. Every search tool takes filters and limits, so ask for a narrower time range or service.

Still stuck? Email info@fixter.dev and one of our founders will help you fix your problem asap.

The docs are readable by agents too: point yours at /llms.txt for how Fixter works, while the MCP server serves your live data.