BaseMouse

capability matrix verified in production

Point BaseMouse at a folder of your docs. It reads them, keeps every version, and labels them so they're easy to find. Re-import whenever you like, with no duplicates, and a clear report if any file doesn't go through.

YOUR AGENT ASKS, AND GETS SOURCED ANSWERS

Your agent makes one request and gets back the most relevant passages, each with its source, a fingerprint that proves it hasn't been tampered with, and how recently it was updated. Ready to drop straight into a prompt.

Every change is recorded, and nothing is ever silently overwritten. Even deletions stay on the record. So you can always answer two questions: where did this answer come from, and what changed since?

Register each tool once — Claude Code, Cursor, Codex, Grok, and more. Every one reads the same grounded context, and whatever an agent writes back in one tool is there for an agent in another, even a different vendor. Your docs and your agents' memory move with you; only the live chat stays inside each tool.

BaseMouse is a remote MCP server atPOST https://basemouse.com/mcp. Add it once and every agent reads and writes the same memory.

Omit the header to browse the public demo corpus. Add your Bearer key for your own workspace.

- search— find relevant context
- get_context_pack— an agent-ready bundle
- upsert_document— persist decisions by stable id

claude mcp add --transport http basemouse https://basemouse.com/mcp \ --header "Authorization: Bearer bm_..."

{ "mcpServers": { "basemouse": { "url": "https://basemouse.com/mcp", "headers": { "Authorization": "Bearer bm_..." } } } }

The demo docs need no key. Your documents need one header.

For your engineers: it's a single HTTP call, no SDK to install. The endpoints below show exactly what comes back.

- GET /api/context-pack— cited, checksummed pack
- GET /api/search— ranked matches with scores
- POST /api/documents— your docs in (Bearer bm_…)
- GET /api/documents/:id/history— the full record

curl -s "https://basemouse.com/api/context-pack?q=agent%20context&limit=2"

{ "schema": "basemouse.context_pack.v1", "entryCount": 2, "citations": [ { "label": "[agent-context-engine] …" } ], "entries": [ { "title": "Agent Context Engine", "relevance": { "score": 6, "matchedTerms": ["agent","context"] }, "provenance": { "checksum": "e765f4dce81039b4", "version": 3 } } ] }

Pick a plan → pay → key issued instantly, shown once. Billed through Stripe. BaseMouse never sees your card.

BaseMouse isopen source at its heart. The engine that imports your docs and serves sourced answers is free and MIT-licensed. Run it on your own servers, with nothing phoning home. The paid hosted version just handles the running, billing, and support for you, so you don't have to. See the docs for exactly which features ship today versus what's on the roadmap.

It's a place to store the documents your AI agents read from. You import your docs, your agent asks a question over one API call, and it gets back the most relevant passages with a source on each one. Every change to every document is kept on the record, so you can always check where an answer came from.

How is this different from a vector database or building RAG myself?

A vector database gives you one ingredient. BaseMouse is the finished dish: import, search, citations, tamper-proof fingerprints, and full change history behind a single endpoint. No pipeline to build, no SDK to install, nothing to keep in sync.

Folders of markdown today. Labels in your files are kept and made searchable, and re-importing is always safe: no duplicates, and a clear report if a file doesn't go through. Docs can also stay in sync on their own: a GitHub Action pushes a repo's agent docs on every push, and a small cross-platform CLI syncs a whole projects folder — or watches it and pushes the moment you save. Bringing docs from another source? That's exactly what thedesign partner programis for.

Yes. The core engine is MIT-licensed and self-hostable, with no license key and nothing phoning home. The paid hosted version adds the running, billing, and support. Grab iton GitHub.

One HTTP call, from any language or agent framework. The demo needs no key at all, and your own docs need one header. Seethe API sectionfor the exact request and response. Prefer a native connection? BaseMouse speaks the Model Context Protocol atPOST /mcp, so agents like Claude Code, Cursor, Codex CLI, Gemini CLI, and Grok CLI connect directly — the samesearchandget_context_packtools, plusupsert_documentso an agent canwrite its own decisions back(idempotently — unchanged content writes nothing), same auth and quota.

AI Reasoning Cache & Consensus Layer with 50+ reasoning objects across 14 domains. Cache hit/miss with token savings, failure risk check before execution, cross-model verification, and citation tracking. 11 MCP tools via Streamable HTTP.