gopher MCP server

Gopher

One MCP server that fetches, caches, and digests context for Claude.

pyarchana.github.io/gopher

Gopher is three things that used to be three separate servers:

• fetch: point it at a GitHub repo and get back a clean Markdown digest: full directory tree, plus the contents of the files that actually matter. Filters out binaries, lock files, node_modules, venv, and the rest of the noise, then ranks what's left. No README? It builds one for you.
• cache: persistent memory across conversations, stored as two plain files you can read yourself: a structured context.json and an append-only diary.md.
• digest: reads a conversation transcript, extracts the facts with a local Ollama model, and merges them into the cache.

Everything runs locally over stdio. Nothing leaves your machine except GitHub API calls.

---

Tools

Tool	What it does
fetch_github_repo(repo_url)	Markdown digest of a public repo, tree plus top files
search_context(query, limit?)	Find facts whose key or value contains query
update_context(key, value)	Set a value at a dot path, e.g. projects.gopher.status
log_diary(entry, tag?)	Append a timestamped Markdown entry
digest_transcript(transcript_path)	Extract facts from a transcript, merge into the store

Five more cover reading, deleting and previewing. The full reference lives on the site, which is checked against the server by tests/test_docs.py so the two cannot drift apart.

Reading memory back

Three ways in, in order of how much they cost you:

search_context("deadline")        # you know what, not where
read_context("projects.gopher")   # you know where
read_context()                    # everything, for when the store is small

search_context is a case-insensitive substring match over key paths and values. Hits on the key rank above hits on the value, results are capped at limit (20 by default), and the reply says how many matched in total so you can tell when you are seeing a slice.

No embeddings, no index, no extra dependency. On a store of a few hundred facts substring matching is enough, and something cleverer can wait for evidence that it is not.

---

Install

Requires Python 3.11+ and uv.

git clone https://github.com/pyarchana/gopher.git
cd gopher
uv venv
uv pip install -e ".[dev]"

The digest tools additionally need Ollama running locally with llama3.2 pulled:

ollama pull llama3.2

Fetch and cache work without it.

---

Claude Desktop

Add this to claude_desktop_config.json (%APPDATA%\Claude\ on Windows, ~/Library/Application Support/Claude/ on macOS):

{
  "mcpServers": {
    "gopher": {
      "command": "uv",
      "args": ["run", "--no-sync", "--project", "/absolute/path/to/gopher", "gopher"]
    }
  }
}

Restart Claude Desktop afterwards.

For Claude Code, register it once and it is available in every session:

claude mcp add gopher --scope user -- uv run --no-sync --project /absolute/path/to/gopher gopher

---

Configuration

Settings come from a .env file in the project root, which the server loads on startup:

cp .env.example .env

Put your GITHUB_TOKEN there rather than in claude_desktop_config.json. The config file gets screenshotted, synced between machines and pasted into issues; .env is gitignored and stays put. Anything set in the client config still wins if you prefer that route.

Variable	Default	Purpose
GITHUB_TOKEN	none	Optional. Raises the GitHub rate limit from 60 to 5,000 requests/hour.
GOPHER_DATA_DIR	./data	Where context.json, diary.md, and digest_log.md live.
GOPHER_DIGEST_BUDGET	40000	Maximum characters a repo digest may return. Raise it if your client accepts larger tool results, lower it to save context.
GOPHER_OLLAMA_URL	http://localhost:11434/api/generate	Ollama endpoint.
GOPHER_OLLAMA_MODEL	llama3.2	Model used for fact extraction.

---

Development

uv run pytest
uv run ruff check .
uv run ruff format .

CI runs all three on every push and pull request, against Python 3.11 through 3.14.

---

Changelog

See CHANGELOG.md.

---

License

MIT, see LICENSE.