MCP DocGraph

MCP server for building knowledge graphs from documents.

Ingest PDF/PPTX/DOCX files, transcribe to Markdown, bootstrap domain-specific entity patterns via LLM, and build a JSONL knowledge graph with RAG chunks — all configurable per project.

Quick Start

pipx install .

Add to your MCP client config:

{
  "mcpServers": {
    "docgraph": {
      "command": "python",
      "args": ["-m", "mcp_docgraph.server"]
    }
  }
}

Set LLM credentials:

export DOCGRAPH_LLM_API_KEY="your-key"
export DOCGRAPH_LLM_PROVIDER="anthropic"  # or "openai"

Related MCP server: pymupdf4llm-mcp

Tools

Workflow

1. Place documents in a folder

2. ingest_files → transcribed Markdown

3. bootstrap_config → proposed config.yaml (review, edit if needed)

4. build_graph → JSONL nodes, edges, chunks

5. validate_graph → check integrity

6. query_graph → ask questions

Design Spec

See docs/superpowers/specs/