> ## Documentation Index
> Fetch the complete documentation index at: https://laminar.sh/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP Server

> Connect MCP clients to Laminar to query traces with SQL and summarize trace context.

Laminar ships an MCP (Model Context Protocol) server so AI assistants can query trace data directly. These are **tools your agent can call**. You ask questions in the chat, the agent invokes the Laminar tools under the hood, and you get answers grounded in your trace data. This closes the dev loop: the agent that wrote the code can pull the exact traces it produced, investigate failures, and propose fixes without context switching.

## Available Tools

* `query_laminar_sql`\
  Run read-only ClickHouse SQL across Laminar data and get JSON back. Use this to find trace IDs, filter by time/session/status, or pull raw span fields (`input`, `output`, `attributes`).
* `get_trace_context`\
  Get an LLM-optimized summary of a single trace: span tree, timings, LLM inputs/outputs, and errors. Use this to quickly understand what happened in a specific run once you have a trace ID.

<Note>
  The SQL tool is read-only (`SELECT` only) and automatically scoped to your project. If trace context truncates long inputs or outputs, use SQL to fetch full fields.
</Note>

## Connect Your Client

First, grab a project API key from the Laminar dashboard. Then connect your MCP client.

<Tabs items={['Claude Code', 'Cursor', 'Codex']}>
  <Tab title="Claude Code">
    ```bash theme={null}
    # Laminar Cloud
    claude mcp add --transport http laminar https://api.lmnr.ai/v1/mcp \
      --header "Authorization: Bearer <YOUR_PROJECT_API_KEY>"

    # Local app-server on port 8000
    claude mcp add --transport http laminar http://localhost:8000/v1/mcp \
      --header "Authorization: Bearer <YOUR_PROJECT_API_KEY>"
    ```
  </Tab>

  <Tab title="Cursor">
    ```json theme={null}
    {
      "mcpServers": {
        "laminar": {
          "type": "http",
          "url": "https://api.lmnr.ai/v1/mcp",
          "headers": {
            "Authorization": "Bearer <YOUR_PROJECT_API_KEY>"
          }
        }
      }
    }
    ```

    To use a local server, change `url` to `http://localhost:8000/v1/mcp`.
  </Tab>

  <Tab title="Codex">
    ```toml theme={null}
    [mcp_servers.laminar]
    url = "https://api.lmnr.ai/v1/mcp"
    bearer_token_env_var = "LMNR_PROJECT_API_KEY"
    ```

    If you want a static token (not an env var), you can set headers directly:

    ```toml theme={null}
    [mcp_servers.laminar]
    url = "https://api.lmnr.ai/v1/mcp"

    [mcp_servers.laminar.http_headers]
    Authorization = "Bearer <YOUR_PROJECT_API_KEY>"
    ```

    For local app-server:

    ```toml theme={null}
    [mcp_servers.laminar]
    url = "http://localhost:8000/v1/mcp"
    bearer_token_env_var = "LMNR_PROJECT_API_KEY"
    ```
  </Tab>
</Tabs>

For self-hosted deployments, replace the base URL with your instance (for example, `https://laminar.example.com/v1/mcp`).

## Example Questions You Can Ask Your LLM

* “What failed in the most recent run for my `dev-claude` session?”
* “Find the latest error trace and summarize the root cause.”
* “Which tool call failed first in the last 20 minutes?”
* “Show me the spans where the model returned an empty response today.”
* “List the 5 slowest traces from the last hour and summarize the slowest one.”
* “Find traces where `status = 'error'` and the top span name contains `checkout`.”
* “Compare the token usage between the last two successful runs.”