Errors and performance
list_exceptions, get_exception, list_endpoints, endpoints_chart, and friends — the same queries the dashboard runs.
Specialized · MCP Server
Production context, over MCP.
Point any MCP client at your Traceway instance. It logs in over OAuth and starts querying your exceptions, logs, latency, and metrics, directly.
Service facts
One protocol, scoped to your instance
The connection, by the numbers.
- 16tools, schema-validated
- Each wraps one API call and declares an output schema; only the two archive tools mutate anything.
- 15 minaccess token life
- Short-lived tokens plus a rotating refresh token; replaying a spent refresh revokes the whole token family.
- 2transports
- Streamable HTTP at /mcp with OAuth, or stdio via traceway mcp on your local login.
Your data, your instance
The server runs where the telemetry lives.
Self-hosted or Cloud, every backend serves /mcp and the OAuth issuer is derived from the request — it works behind a reverse proxy with no extra configuration. MIT licensed, same tools either way.
claude mcp add --transport http traceway https://cloud.tracewayapp.com/mcpFirst connect opens your browser on the consent page — no token to paste.
- OAuth
- RFC 7591 + PKCEThe client registers itself, you approve on your own instance's consent page — no admin screen, no client secret.
- CI fallback
- Personal access tokenNon-OAuth clients send a PAT on the Authorization header instead.
- Local
- traceway mcpThe CLI doubles as a stdio MCP server, reusing your CLI session and current project.
- Headless
- TRACEWAY_URL + TRACEWAY_TOKENContainers skip the session entirely; the server exits with instructions if it can't answer.
On this line
Sixteen tools. Read-only by default.
Errors, performance, signals, and projects — each tool wraps one API call and returns validated structured content.
Signals
query_logs, query_metrics, get_task, get_ai_trace, get_session, and get_trace cover the rest of the telemetry.
Prompts and resources
The same investigation playbooks the Traceway skill ships, exposed as MCP prompts (debug_issue, investigate_performance) and knowledge resources.
Scoped to the approver
Tokens are scoped to the user who approved the connection; if you can't see a project in the dashboard, a tool can't either.
Safe by default
Read-only unless you say otherwise.
Every read tool is annotated read-only. The two archive tools are marked mutating and instruct agents to call them solely on explicit request — investigating an error never touches it.
Next stop
Give your agent a live connection.
Connect an MCP client to your instance and let it investigate the next issue.