MCP

Ask your telemetry questions in plain text

Uptrace ships with a built-in Model Context Protocol server. Point Claude Code, Cursor, or Codex at your project and query spans, traces, alerts, and dashboards directly from your AI assistant, no dashboard clicking, no query language to learn first.

Connect your AI assistant Schedule a demo

14-day free trial · No credit card required

Connect in one command

Grab a project ID and a user API token from Uptrace, then point any MCP-compatible client at your project. No SDK, no separate integration to build.

One command adds the server to Claude Code
Or paste the same endpoint into any mcp.json-compatible client
Project ID and token come pre-filled from the MCP page in Uptrace
Works with Claude, Cursor, Continue, and other MCP clients

Add the MCP server

Requires an API token — pick one above or create a new one.

claude mcp add --transport http uptrace \

https://<your-uptrace-host>/mcp/<project_id> \

--header "Authorization: Bearer <your-user-token>"

Read spans and traces without writing UQL

Four tools cover everything from one slow request to aggregated error rates across a service, using the same search syntax and system filters as the Uptrace UI.

Inspect individual spans or aggregate them into groups with error rate and percentiles
Find whole traces with correlated sub-queries across multiple span types
Full search syntax: word, phrase, AND/OR/exclude, regex, scoped attribute search
System filters like httpserver:all, db:postgresql, and log:error

list_spans

Individual span records matching a query

list_span_groups

Spans aggregated with count, error rate, percentiles

list_traces

Whole traces via correlated sub-queries

list_trace_groups

Traces aggregated across multiple span types

Alerts and metrics as callable tools

An assistant can check what is currently firing, inspect a specific alert, and explore any metric, its attributes, and their values, without leaving the chat.

List alerts filtered by status, priority, type, or monitor
Get a single alert with full event history and assignees
List monitor rules with type, state, query, and channels
Discover metrics, their instrument type, unit, and attribute keys

list_alerts

Filter by status, priority, type, or monitor

get_alert

Full event history and assignees for one alert

list_monitors

Monitor rules, state, query, and channels

explore_metrics

Name, instrument, unit, and attributes

Create and update dashboards from a prompt

Dashboard tools use the same YAML schema as the UI import/export feature, so an assistant can scaffold a dashboard and you can still edit it by hand afterward.

List dashboards and tags already in the project
Export any dashboard as YAML, or start from a built-in template
Create or replace a dashboard definition in one call
Same YAML schema as manual dashboard import and export

get_dashboard_yaml

Export an existing dashboard as YAML

create_dashboard_from_template

Create a dashboard from a YAML definition

update_dashboard_from_template

Replace a dashboard with new YAML

list_dashboard_templates

Browse built-in starting points

Better together

Each signal is more powerful when correlated with the others. Uptrace stores traces, logs, and metrics in one place, no tool-switching, no data gaps.

Distributed tracing

Find the exact span that broke your request, across every service and database call.

Log management

Search structured logs at any scale and jump from any span to its logs in one click.

Metrics

PromQL-compatible metrics with no cardinality limits and automatic dashboards.

Common questions, clear answers

What is the Uptrace MCP server?

A built-in Model Context Protocol server that exposes your telemetry as MCP tools and resources. It lets AI assistants like Claude, Cursor, and Continue query spans, traces, metrics, and alerts, and manage dashboards, directly from a chat session instead of the Uptrace UI.

How do I authenticate MCP requests?

Each request carries a user API token in the Authorization header. The token is user-scoped, not project-scoped: it grants the same permissions the user has across their projects, and the project ID in the URL selects which project the tools operate on.

Can an AI assistant modify my data, or only read it?

It depends on the tool. Telemetry tools (spans, traces, metrics, alerts, monitors) are read-only. Dashboard tools can create, update, and delete dashboards, since that is a safe, reversible action tracked like any other dashboard change.

Does the MCP server work if my organization uses SSO?

User API tokens do not work with SSO, because SSO requires browser redirects that an MCP client cannot perform. If your organization uses SSO, create a separate user account with static credentials and grant it access to the relevant projects.

Which AI assistants can connect?

Any MCP-compatible client: Claude Code and Claude Desktop, Cursor, Continue, and others. The MCP project page in Uptrace shows ready-to-paste claude mcp add and codex mcp add commands with your project ID and token pre-filled.

Ready to let your AI assistant read the room?

Deploy in minutes with Docker. Open source. Fixed pricing. Your data stays on your infrastructure.

Connect your AI assistant

14-day free trial · No credit card required

Learn more

Documentation Pricing Schedule a demo