@infoinlet/mcp-knowledge-rag

MCP server for a permission-aware RAG knowledge base — a thin client over the hosted knowledge-rag API. Let your agent ingest documents and ask grounded, cited questions, with per-document access control.

Free to install; calls a paid hosted API. Start a free trial / get a key at infoinlet.com. No model/embedding keys needed — the server is just a client; chunking, embeddings, retrieval, and answer composition all run on the hosted service.

Tools

| Tool | What it does | |---|---| | ingest | Chunk + embed + index a document (dense + lexical lanes). Re-ingesting the same externalId replaces it. Optional permissionSet. | | search | Hybrid (dense + lexical) retrieval, filtered by caller permissions. Returns top chunks. | | answer | Grounded answer with [n] citations; refuses outside knowledge. | | delete_document | Remove all chunks of a document from both lanes. | | workspace_stats | Document/chunk counts + last index time. | | usage | Your key's usage/quota (trial vs. unlimited). |

Configure

KNOWLEDGE_RAG_API_KEY=sk-...                       # required
KNOWLEDGE_RAG_BASE_URL=https://api-rag.knowlez.com # optional override

Claude Desktop / any MCP client

{
  "mcpServers": {
    "knowledge-rag": {
      "command": "npx",
      "args": ["-y", "@infoinlet/mcp-knowledge-rag"],
      "env": { "KNOWLEDGE_RAG_API_KEY": "sk-..." }
    }
  }
}

Notes

• Multi-tenant by workspaceId; documents are gated by permissionSet and retrieval is filtered by caller permissions.
• Errors are surfaced plainly: 401 (bad/missing key), 402 (trial quota used up — subscribe), 429 (rate limited).