Skip to content
MCP server

Link

Local Markdown-based memory shared across your coding agents, with review-gated writes

Runs locallyStatus and history
Install
Pricing
Open source
Platforms
macOS, CLI, self-hosted
Runs
Locally
Version
5.0.0
Listing updated
today

Similar MCP servers

About Link

Link is an open-source MCP server for knowledge and memory. It runs locally on your machine and is started by your AI client.

Link keeps agent memory as plain Markdown files on your own machine instead of an embedding database. Agents can only propose memories; nothing becomes durable until you approve it, and the README stresses that no language model sits in the write or recall path, so ingestion and retrieval are deterministic.

Beyond recall, a reviewed memory can carry a rule such as asking before a force push or denying writes to a migrations folder, checked before a tool call runs. Memories that name files, scripts or functions are compared against the repository and flagged when the code has moved on, and a receipt command lists which memories reached each session.

The README reports benchmark results, including 84.8% on LoCoMo question answering and 99.4% evidence retrieval on LongMemEval; these are the author's own figures. It targets developers who use several agents such as Codex, Claude, Cursor, Kiro and Copilot and want one shared memory.

Install

Run in a terminal

claude mcp add link -- uvx link-mcp

Add to claude_desktop_config.json

{
  "mcpServers": {
    "link": {
      "command": "uvx",
      "args": [
        "link-mcp"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "link": {
      "command": "uvx",
      "args": [
        "link-mcp"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "link": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "link-mcp"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "link": {
      "command": "uvx",
      "args": [
        "link-mcp"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "link": {
      "command": "uvx",
      "args": [
        "link-mcp"
      ]
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

What it does

Review-gated memory
Agents propose memories and you approve them before they persist.
Plain Markdown storage
Every memory and wiki page is a file you can open, grep or git-diff.
Enforced rules
Approved memories can ask or deny tool calls before they run.
Stale-memory checks
References to files and functions are checked against the repository.
Session receipts
See which memories were given to each session.
Compact recall packets
Query results are bounded by a token budget with provenance.

Good for

  • Share project decisions between Claude Code and Cursor
  • Stop agents repeating a forbidden command like a force push
  • Import existing Claude Code memory into one store
  • Audit what context an agent session was given

Pricing

Open source

Open-source; install via Homebrew or pip

Link FAQ

How do I install it?

Run brew install gowtham0992/link/link or pip install link-mcp, then lnk setup to wire up the agents you have.

Does memory leave my machine?

The README says it is local and that CI blocks outbound network code in the runtime; optional semantic models load offline after one explicit setup.

Does it use an LLM to write memories?

No. The README says there is no model in the memory write path.

What is the transport?

stdio, from the PyPI package link-mcp.

Is Link free?

Yes. Link is open source, so you can use and self-host it for free.

How do I add Link to Claude Code?

Run this command in your terminal: claude mcp add link -- uvx link-mcp

What are the best alternatives to Link?

The closest alternatives to Link are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.

Where this comes from

  • Description: written from the project's README.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers