# Fixter

> Fixter is monitoring for teams that build with coding agents: logs, traces, and
> metrics, with alerting, error tracking, and AI investigations, all of it exposed to
> your agent as MCP tools. This is the marketing site. The documentation lives at
> https://docs.fixter.dev and is written for agents too.

## Set up Fixter

If someone pasted "Read fixter.dev/llms.txt and set up Fixter." into you, do this, in
order. It works from any coding agent; where a step differs by client, every client is
listed. The full walkthrough is https://docs.fixter.dev/getting-started/onboarding.md

1. Connect the MCP server: https://mcp.fixter.dev/mcp (Streamable HTTP). It signs
   the person in with OAuth in the browser on first connection, so no API key is
   needed when a person is present. Add it the way your client expects:

   - Claude Code, in a terminal:

         claude mcp add --transport http fixter https://mcp.fixter.dev/mcp

     or commit `.mcp.json` at the repository root so the whole team gets it:

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

     Sign in: the person runs `/mcp` inside Claude Code.

   - Cursor, in `.cursor/mcp.json` (one project) or `~/.cursor/mcp.json` (all):

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

     Sign in: the person enables it under Cursor Settings > MCP; the browser opens
     on first connection.

   - Codex CLI, in a terminal:

         codex mcp add fixter --url https://mcp.fixter.dev/mcp

     or in `~/.codex/config.toml` (or `.codex/config.toml` in the project):

         [mcp_servers.fixter]
         url = "https://mcp.fixter.dev/mcp"

     Sign in: the person runs `codex mcp login fixter`.

   - Any other client: configure a remote MCP server at that endpoint with
     Streamable HTTP transport and leave auth to the client; Fixter serves the
     standard MCP OAuth discovery documents, so the client walks the browser
     sign-in on its own.

   An agent cannot complete the sign-in itself, so ask the person to do it before
   going on. Signing in is what lets you fetch ingestion credentials. Per-client
   detail: https://docs.fixter.dev/setup/mcp.md

2. Get telemetry flowing. Pick the row that matches where you are running.

   - Claude Code: use the plugin rather than wiring OpenTelemetry by hand. These are
     slash commands the person pastes into Claude Code, one per paste; an agent
     cannot run them, so show them to the person:

         /plugin marketplace add fixter-dev/fixter-skills
         /plugin install fixter@fixter-skills
         /reload-plugins
         /fixter:onboard

     `/fixter:onboard` detects the stack, adds the OpenTelemetry packages and
     environment variables, provisions the API key as a one-time link, then reviews
     logging and sets up alerts. It resumes across sessions.

   - Cursor, Codex, or any other coding agent connected over MCP: follow
     https://docs.fixter.dev/setup/opentelemetry.md for the instrumentation (it
     covers Kubernetes, an existing Collector, and direct export per language).
     For the credentials, skip that page's "create an API key" step and call the
     MCP tool `get_ingestion_credentials` instead: it returns the ingestion
     endpoint and a single-use link (expires in 24 hours) that the person opens to
     reveal the key. The key never comes back through the tool, so ask the person
     for it. In a header the value is `Authorization=Bearer <key>`; in a `.env`
     file write the literal key.

   - No MCP connection, or no agent at all: the person creates a key at
     https://app.fixter.dev/settings/api-keys and sets four environment variables
     on the service (one OTEL_SERVICE_NAME per service, e.g. `checkout-service`):

         OTEL_SERVICE_NAME=<service-name>
         OTEL_EXPORTER_OTLP_ENDPOINT=https://ingest.fixter.dev
         OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
         OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer $FIXTER_API_KEY"

3. Deploy (or ask the person to), wait two or three minutes, then confirm that
   logs, traces, and metrics each arrive. Ask for the three signals by name: traces
   can show up while logs never do. Over MCP, use the `logs` and `spans` tools
   filtered by service name (https://docs.fixter.dev/reference/mcp-tools.md lists
   every tool), or open https://app.fixter.dev/logs

4. Next: connect Slack so alerts land in a channel and get investigated
   automatically (https://docs.fixter.dev/setup/slack.md), and install the GitHub
   App so investigations can open pull requests (https://docs.fixter.dev/setup/github.md).
   Both are installs the person completes in a browser. Alert thresholds calibrate
   against history, so alert setup often defers until a day of logs has
   accumulated. Re-run it then.

## Documentation

- [Docs index for agents](https://docs.fixter.dev/llms.txt): every documentation page with a one-line description.
- [Complete documentation in one file](https://docs.fixter.dev/llms-full.txt): the whole corpus in a single fetch.
- Any docs page is available as raw markdown: append `.md` to its URL.
- [Onboarding](https://docs.fixter.dev/getting-started/onboarding.md): from nothing to logs arriving, then alerts in Slack and investigations.
- [MCP](https://docs.fixter.dev/setup/mcp.md): connecting Claude Code, Cursor, or any MCP client; API keys for headless agents.
- [MCP tools](https://docs.fixter.dev/reference/mcp-tools.md): every tool Fixter exposes, grouped by product area.
- [Send data with OpenTelemetry](https://docs.fixter.dev/setup/opentelemetry.md): per-language instrumentation, log bridges, and collector paths.
- [Claude Code plugin](https://docs.fixter.dev/setup/claude-code-plugin.md): the onboarding command and the five skills.
- [QuerySQL](https://docs.fixter.dev/reference/querysql.md): the SQL dialect for querying logs, spans, and metrics.

## What Fixter does

- Monitoring: stores logs, traces, and metrics sent over standard OpenTelemetry, from any language or platform, plus Kubernetes and AWS CloudWatch collectors.
- Alerts: rules written in plain language, plus built-in detectors for error bursts and anomalies that need no configuration.
- Issues: exceptions parsed out of the telemetry you already send, grouped, tracked, and investigated automatically.
- Investigations: an AI agent reads your telemetry when an alert fires, finds the root cause, cites the exact logs and traces, and replies in the Slack thread.
- Debugging: query production from your coding agent over MCP, in the app, or in SQL.
- Dashboards: ask your agent for a dashboard and get a stable, shareable link it updates in place.
- Uptime: scheduled probes against HTTP endpoints and MCP servers, with assertions and latency tracking.

## This site

- [Home](https://fixter.dev/): what Fixter is and the one-prompt setup.
- [Product](https://fixter.dev/product): monitoring, issues, debugging from your coding agent, alerts, AI investigations, and uptime monitoring.
- [Pricing](https://fixter.dev/pricing): free forever with 10 GB a month and no card; paid from €20 a month; unlimited users on every plan.
- [AI SRE](https://fixter.dev/ai-sre): how investigations work, from alert to verdict to the fix in Claude Code.
- [Blog](https://fixter.dev/blog): articles on monitoring, observability, and building for AI agents.
- [App](https://app.fixter.dev/): sign in.
- Contact: info@fixter.dev