53 lines
2.0 KiB
Markdown
53 lines
2.0 KiB
Markdown
# LightRAG MCP Server
|
|
|
|
An MCP server exposing the LightRAG Server API as tools, resources, and prompts for coding agents.
|
|
|
|
## Features
|
|
|
|
- Retrieval tools: `query_data`, `query`, `query_stream`, `query_stream_chunks`
|
|
- Ingestion tools: `ingest_text`, `ingest_texts`, `ingest_file`, `ingest_files`, `upload_document`
|
|
- Freshness tools: `scan_documents`, `scan_and_wait`, `pipeline_status`, `wait_for_idle`, `track_status`
|
|
- Memory tool: `ingest_memory` for lessons, preferences, decisions, structures, functions, relationships
|
|
- Graph tools: entity/relation CRUD, entity existence check, label search, graph export
|
|
- Health tool: `health`
|
|
- Macro tool: `refresh_and_query` (scan -> wait idle -> query_data -> query)
|
|
- Resources: health, pipeline status, documents, graph, labels (list + popular), status counts
|
|
- Prompts: evidence-first answering, refresh-then-query, record project memory
|
|
|
|
## Quickstart
|
|
|
|
```bash
|
|
# Create a venv and install
|
|
python -m venv .venv
|
|
. .venv/bin/activate
|
|
pip install -e .
|
|
|
|
# Run the MCP server (streamable HTTP)
|
|
export MCP_TRANSPORT=streamable-http
|
|
export MCP_HOST=127.0.0.1
|
|
export MCP_PORT=8000
|
|
export LIGHTRAG_BASE_URL=http://127.0.0.1:9621
|
|
lightrag-mcp
|
|
|
|
# Smoke test (health + optional retrieval)
|
|
lightrag-mcp-smoke --query "What is this project?" --format pretty
|
|
```
|
|
|
|
## Configuration
|
|
|
|
- `LIGHTRAG_BASE_URL` (default `http://127.0.0.1:9621`)
|
|
- `LIGHTRAG_TIMEOUT_S` (default `60`)
|
|
- `LIGHTRAG_POLL_INTERVAL_S` (default `1`)
|
|
- `LIGHTRAG_POLL_TIMEOUT_S` (default `120`)
|
|
- `MCP_TRANSPORT` (default `streamable-http`)
|
|
- `MCP_HOST` (default `127.0.0.1`)
|
|
- `MCP_PORT` (default `8000`)
|
|
- `MCP_SERVER_NAME` (default `LightRAG MCP`)
|
|
|
|
## Notes
|
|
|
|
- `query_stream` collects the streaming response and returns it as a single string.
|
|
- `query_stream_chunks` returns chunked output and reports progress to clients that support progress events.
|
|
- `refresh_and_query` is a convenience macro for evidence-first workflows.
|
|
- `ingest_file(s)` chunk local files and store them with `file_source` references.
|