Graphlit MCP Server

npx -y graphlit-mcp-server

{
  "mcpServers": {
    "graphlit-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "graphlit-mcp-server"
      ]
    }
  }
}

The server treats a Graphlit project as one searchable knowledge base sitting across the tools a team already uses, and the ingestion side does real work on the way in: PDF, DOCX, PPTX and HTML are extracted to Markdown, audio and video are transcribed, so what lands in the index is text a model can actually retrieve against. Web crawling and web search are built in as tools, which removes the usual second integration with a scraping service. The less obvious half is publishing, which runs the other direction — a retrieved answer can be rendered as ElevenLabs audio or an OpenAI-generated image. Retrieval covers contents, collections, feeds and conversations, plus similar-image search and visual image description.