knowledge-cache

MCP server for caching technical documentation, API references, and code patterns in SQLite + FTS5. Reduces repeated web lookups by storing knowledge locally with tag-based categorization and full-text search.

How it works

Agent -> knowledge_find("navmesh setdestination", ["unity"])
            ├── hit → returns cached content (avoids web fetch)
            └── miss → fetch via context7 → knowledge_store(content, tags, url)

Tools

Storage

SQLite with FTS5 for full-text search. Schema:

• knowledge — entries with content, source URL, timestamps
• knowledge_fts — FTS5 virtual table (porter stemmer, prefix matching)
• knowledge_tags — category tags per entry (NOCASE, multi-tag filter)

Default database path: ./.knowledge/knowledge.db (configurable via --db-path).

Setup

Install

cd knowledge-cache
uv add mcp

MCP config (per project .mcp.json)

{
  "mcpServers": {
    "knowledge-cache": {
      "command": "uv",
      "args": [
        "run",
        "--directory", "/path/to/knowledge-cache",
        "server.py",
        "--db-path", "/path/to/project/.knowledge/knowledge.db"
      ]
    }
  }
}

Dependencies

Zero (stdlib: sqlite3, os, argparse). Requires mcp package for the MCP server transport. FTS5 is built into Python's SQLite on Windows.