Skip to main content
Glama

engram-mcp

PyPI License: MIT

MCP plugin that connects Cursor, Claude Code, Claude Desktop, and any other MCP-speaking host to your Engram workspace — the indexed knowledge of your GitHub repos and Notion pages.

The plugin is a thin client: no local database, no embedding model, no vector store. It forwards tool calls to the Engram backend over HTTPS and renders the results for your LLM.

Install

pip install engram-mcp
# or
uv tool install engram-mcp

Related MCP server: PLUR

Configure your MCP client

You need two things:

  1. Your Engram backend URL — https://your-engram.example.com in production, or http://localhost:8000 for local development.

  2. An Engram MCP token — mint one from your Engram web app under Settings → MCP Tokens. Tokens are scoped to your user and expire after 90 days.

Claude Desktop / Claude Code

Add to claude_desktop_config.json (or the equivalent Claude Code config):

{
  "mcpServers": {
    "engram": {
      "command": "engram-mcp",
      "env": {
        "ENGRAM_BASE_URL": "https://your-engram.example.com",
        "ENGRAM_TOKEN": "eng_mcp_..."
      }
    }
  }
}

Cursor

In Cursor settings → MCP → Add new server:

{
  "engram": {
    "command": "engram-mcp",
    "env": {
      "ENGRAM_BASE_URL": "https://your-engram.example.com",
      "ENGRAM_TOKEN": "eng_mcp_..."
    }
  }
}

Codex CLI

# ~/.codex/config.toml
[mcp_servers.engram]
command = "engram-mcp"
env = { ENGRAM_BASE_URL = "https://your-engram.example.com", ENGRAM_TOKEN = "eng_mcp_..." }

Tools exposed

Tool

What it does

search_knowledge

Ranked search across your indexed GitHub + Notion content. Supports source filter and intent shaping (explain / generate / question).

cite

Locators only (source, path, line range, URL) for an answer the LLM has already drafted — cheap on context.

fetch_document

Full document fetch by id, optionally with surrounding context lines. Use after search_knowledge surfaces something interesting.

How it works

┌───────────────────┐  stdio MCP   ┌───────────────────┐  HTTPS   ┌──────────────────┐
│ Cursor / Claude   │ ───────────► │ engram-mcp        │ ───────► │ Engram backend   │
│ Code / Desktop    │              │ (this package)    │  bearer  │ (indexing, RAG,  │
└───────────────────┘              └───────────────────┘   token  │  cross-val)      │
                                                                  └──────────────────┘

The plugin holds no state. Your token authenticates each request to the Engram backend, which resolves it to your user + workspace, runs retrieval against the indexed corpus, and returns ranked chunks. The plugin formats those chunks as markdown for the host LLM.

Auth

v0.x uses paste tokens — generate one in the Engram web app, paste into the env var. Tokens are stored hashed server-side and you can revoke them from the same settings page.

v1.x will add OAuth 2.1 (device flow + Dynamic Client Registration) so you can add Engram as a claude.ai connector with zero copy-paste. Paste tokens will keep working.

Development

git clone https://github.com/hidden-ones-dev/engram-mcp
cd engram-mcp
pip install -e ".[dev]"
pytest

Point at a local Engram backend:

export ENGRAM_BASE_URL=http://localhost:8000
export ENGRAM_TOKEN=eng_mcp_...
python -m engram_mcp   # speaks stdio MCP; pipe from a host or test harness

License

MIT — see LICENSE.

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    D
    maintenance
    Obsidian-backed knowledge graph with semantic search, entity extraction, and cross-session memory. 11 MCP tools. Works with Claude Code, Cursor, Windsurf, and any MCP-compatible editor.
    37 npm
    1
    MIT
  • A
    license
    B
    quality
    A
    maintenance
    Open, local-first engram memory for AI agents: read, correct, and delete what your agent remembers. Works across Claude Code, Hermes, OpenClaw, and Cursor over MCP.
    39
    1,858,293 npm
    267
    Apache 2.0
  • F
    license
    C
    quality
    D
    maintenance
    Provides RAG (Retrieval-Augmented Generation) capabilities via Contextual AI, enabling query processing and context-aware responses with citations. Integrates with MCP clients like Cursor IDE and Claude Desktop.
    1
    -