Skip to main content
Glama
squatch-codebook-mcp

squatch-codebook-mcp

by squatch-c-c

squatch-codebook-mcp

Reason over your own local notes with a hyperdimensional (FHRR) codebook — query, related, neighbors, walk, analogy. No LLM in the loop, nothing leaves your machine. An MCP server, so it works in Claude Desktop (Chat) and Claude Code.

Run it (zero install — uvx fetches + runs)

uvx --from git+https://github.com/squatch-c-c/squatch-codebook-mcp squatch-codebook-mcp

Requires uv (curl -LsSf https://astral.sh/uv/install.sh | sh).

Related MCP server: MCP Associative Memory Server

Add to Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

{
  "mcpServers": {
    "codebook": { "command": "uvx", "args": ["--from", "git+https://github.com/squatch-c-c/squatch-codebook-mcp", "squatch-codebook-mcp"] }
  }
}

Restart Claude. Now in Chat or Code you can ask things like "use codebook to find notes in ~/notes related to hyperdimensional computing" and it answers geometrically over your local files.

Tools

Tool

What it does

codebook_build(notes_dir, glob="*.md")

Index a folder; list its notes (run first).

codebook_query(notes_dir, text, …)

Notes most related to a phrase.

codebook_related(notes_dir, note_id, …)

Related notes (geometric + link recall).

codebook_neighbors(notes_dir, note_id, …)

Nearest notes by cosine similarity.

codebook_walk(notes_dir, start_id, steps=4)

Reasoning walk along the LINKS relation.

codebook_analogy(notes_dir, a, b, c, …)

a:b :: c:? over your notes.

All paths are local; every tool reads only the directory you point it at. Part of the Squatch working system.

Available Tools

6 tools
codebook_analogyC

Analogy a:b :: c:? over your notes (VSA transform), returns the best completions.

Inputs: notes_dir (required), a (required), b (required), c (required), +2 more. Returns: result

codebook_buildA

Index a local notes folder and return its codebook contents (id + title per note). Run this first to see what's in a directory. notes_dir is a local path; glob picks files.

Inputs: notes_dir (required), glob.

codebook_neighborsC

Nearest notes to a given note by pure cosine similarity in the hypervector space.

Inputs: notes_dir (required), note_id (required), glob, k. Returns: result

codebook_queryB

Find the notes most related to a phrase, geometrically (no keyword match, no LLM).

Inputs: notes_dir (required), text (required), glob, k. Returns: result

codebook_walkB

Reasoning walk: from a note, repeatedly predict the next via the LINKS relation.

Inputs: notes_dir (required), start_id (required), glob, steps. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.4/5.0

    Scored across 6 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose: indexing, analogy, nearest neighbors, query by phrase, related notes, and reasoning walk. No overlap or ambiguity.

    Naming Consistency5/5

    All tools follow a consistent pattern: 'codebook_' prefix followed by a descriptive noun (analogy, build, neighbors, query, related, walk). No mixing of styles.

    Tool Count5/5

    With 6 tools, the server is well-scoped for its purpose of note indexing and geometric reasoning. Not too few or too many.

    Completeness4/5

    Covers indexing, various retrieval methods, analogy, and reasoning. Minor gap: no incremental note addition or update; only reindexing entire folders, which could be a limitation.

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      B
      quality
      D
      maintenance
      Enables indexing and retrieving notes with full-text search using SQLite, plus building knowledge graphs to find relationships between concepts. Supports natural language note management, tagging, and semantic connections.
      16
      -
    • A
      license
      A
      quality
      D
      maintenance
      Turns a local folder of notes and documents into a searchable knowledge base for AI assistants via MCP, enabling semantic search, reading, and adding notes entirely on-device.
      4
      9
      MIT