Repiscope

What is Repiscope?

Repiscope is a read-only MCP server that gives coding agents (Claude Code, Cursor, or any MCP client) awareness of sibling repositories next to the one they work in — without ever letting the agent modify them. It is built so that safe behaviour is the only behaviour possible: the server exposes zero write tools, structurally preventing any modification to your repos.

How to use Repiscope?

Install via pip install repiscope, then add it to your MCP client with --root pointing to the folder containing your repos (not a repo itself). Optionally hide repos with --exclude repo-a --exclude repo-b. For Claude Desktop there is also a one-click .mcpb bundle (built from mcpb/build.sh). Works with any MCP client; from source, clone the repo and run pip install -e .[dev].

Key features of Repiscope

• Zero repo-write tools – the agent cannot modify your repositories
• Secrets invisible – files containing keys, certs, .env* are filtered out
• Symlinks can't smuggle – files outside the project are invisible
• You define the perimeter via --root and --exclude flags
• Lazy cache refresh – overviews rebuilt only when a repo's commit changes
• Borrowed intelligence – no built-in LLM; calls the agent to write summaries

Use cases of Repiscope

• A coding agent in project A needs to understand project B's API without writing to B
• A personal assistant decides where your time goes by seeing real state of every project
• Any scenario where an agent needs structured, read-only sight of sibling repos while its sandbox stays tight

FAQ from Repiscope

What does Repiscope do that widening the sandbox doesn't?

Widening the sandbox would grant write access too. Repiscope gives the agent structured, read-only sight of every sibling repo through one MCP server, and the sandbox stays tight.

What are the runtime requirements?

Python ≥ 3.11. Developed and tested on macOS; Linux should behave identically, Windows symlink semantics are untested.

Where is data cached and stored?

Overview caches live in ~/.cache/repiscope, never inside your repositories. The store_summary tool writes only to Repiscope's own cache.

What are the known limits of the tools?

Search is a case-insensitive substring match with capped results, no index, no regex. The secret filter works at file level – a password pasted inside an ordinary text file is not scrubbed. There is no manual refresh tool; caches refresh lazily on the next commit.

Does Repiscope require authentication or network access?

No. Repiscope has no LLM of its own, no API key, no model calls, and zero ongoing cost. It runs locally via your MCP client (e.g., STDIO transport).