link-mcp 1.0.2

MCP server for the Link personal knowledge wiki — search, context, and graph traversal

link-mcp

MCP server for the Link personal knowledge wiki. Exposes your wiki as MCP tools — search, query context, and traverse the knowledge graph without reading files directly.

Listed on the official MCP Registry as io.github.gowtham0992/link.

Install

pip install link-mcp

Quick setup (Kiro)

git clone https://github.com/gowtham0992/link.git
bash link/integrations/kiro/install.sh

This installs link-mcp, scaffolds ~/link/, and registers the MCP server in ~/.kiro/settings/mcp.json automatically.

Manual setup (any MCP client)

1. Scaffold your wiki:

git clone https://github.com/gowtham0992/link.git
bash link/integrations/kiro/install.sh   # or claude-code, cursor, codex

2. Add to your MCP client config:

{
  "mcpServers": {
    "link": {
      "command": "python3",
      "args": ["-m", "link_mcp"]
    }
  }
}

Custom wiki path:

{
  "mcpServers": {
    "link": {
      "command": "python3",
      "args": ["-m", "link_mcp", "--wiki", "~/my-wiki/wiki"]
    }
  }
}

Tools

Tool	Description
search_wiki(query, limit?)	Ranked search — title (20pts), alias (8pts), tag (5pts), fulltext (2pts). Returns scores + snippets.
get_context(topic)	Primary tool. Best matching page (full content) + inbound/forward graph links in one call.
get_pages(category?, type?, maturity?)	All pages with metadata. Filter by category, type, or maturity.
get_backlinks(page_name)	Inbound + forward links for a page.
get_graph()	All nodes + edges for graph reasoning.
rebuild_backlinks()	Rebuild _backlinks.json after ingest or lint.

Use get_context for answering questions — one call returns the primary page plus all related pages via graph traversal. Eliminates the token waste of reading index.md every session.

Wiki location

Default: ~/link/wiki/. Override with --wiki /path/to/wiki.

Requirements

• Python 3.10+
• A Link wiki (scaffolded by install.sh)