lerim mcp

Run Lerim's stdio MCP server for compatible agent clients.

Most users do not start this command by hand. lerim connect <agent> --mode mcp writes client config that launches it automatically.

Usage

lerim mcp

Generic MCP client config:

{
  "mcpServers": {
    "lerim": {
      "command": "/absolute/path/to/python",
      "args": ["-m", "lerim.mcp_server"]
    }
  }
}

lerim connect writes this using the Python executable that is running Lerim, so external MCP clients do not need a lerim executable on their PATH.

Tools

Tool	Purpose
lerim_context_brief	Read or refresh startup context for a project.
lerim_context_answer	Ask a grounded question over stored context records.
lerim_context_search	Retrieve compact records for a query.
lerim_records_list	List records deterministically with filters.
lerim_trace_submit	Submit a completed session transcript for normal extraction.
lerim_ingest_status	Inspect runtime, queue, ingest, and project status.

lerim_trace_submit accepts JSONL, JSON arrays, JSON objects with messages, events, trace, steps, or items, and plain text transcripts. Wrapper metadata can provide session_id, and identical normalized content is skipped for the same session unless force=true is passed.

Failed submissions are retryable. The tool response includes submitted_trace_path, submission_manifest_path, and retry_command; the same saved metadata can be inspected with lerim trace submissions and retried with lerim trace retry <submitted_trace_path>.

Notes

MCP is the access layer. Lerim's durable context still lives in the global SQLite context store, and extraction still runs through Lerim's normal DSPy extraction pipeline.

Think of MCP as two tool groups:

• Read tools: lerim_context_brief, lerim_context_answer, lerim_context_search, and lerim_records_list retrieve context that Lerim has already extracted.
• Submit tool: lerim_trace_submit is an explicit import path for one completed session. It is not automatic background capture.

When calling lerim_trace_submit, pass the scope explicitly. For repository work, use scope_type="project" and scope="/absolute/path/to/repo". For business workflows, use a domain or workspace scope such as scope_type="domain" and scope="support-ops". Relying on the MCP server's current working directory can put submitted context under the wrong scope in clients that launch MCP servers from a generic directory.