Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@50 First Tapes MCP Server query notes about vector search"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
An open, local-first knowledge engine. Your notes are plain markdown and git — the tapes; your AI reads and writes them behind governance gates. No lock-in: git clone and walk away.
Apache-2.0 · OKF-compatible · TypeScript. The v1 protocol verbs work today. This repo is the engine and carries no personal data.
Knowledge that outlives any one tool or chat, that an AI agent can read and write safely, and that you can build products on — a personal vault, a team wiki, an agentic LMS. The field converged on this pattern in 2026 (Karpathy's LLM wiki, Google's Open Knowledge Format). 50 First Tapes adds the hard parts those leave out: trust, governance, multi-agent writes, and a typed schema system.
Related MCP server: brain-mcp
Products & views — your vault · a team wiki · an agentic LMS
Agents & governance — crew + rules as data; gates; audit; the loop runner
Protocol — ingest · query · lint · write · govern (MCP · CLI · library)
Derived index — full-text + vector + graph; rebuilt on demand; never committed
Substrate — markdown + YAML + git; OKF-compatible; the durable floorOne protocol, three shapes — a CLI (tapes), an MCP server, and a library. The five verbs:
Verb | What it does |
| Bring raw material in as a governed, kind-scaffolded note. Gates block; schema gaps are reported, not blocked. |
| Find notes by free text (title-weighted), kind, tag, status, or backlinks. Reads on demand — no committed index. |
| Validate every note against its kind, and check structural health: broken/stale links, orphans, committed blobs, hot-core budget. |
| Write exact content to a note id, through the governance gates. |
| Dry-run the gates on proposed content without writing. |
tapes lint # validate the bundle in the current dir
tapes query "vector search" --kind note # find notes
tapes ingest article.md --kind source # file new material under inbox/
tapes govern wiki/draft draft.md # would this write be accepted?Governance. Writes pass gates before they land: an off-limits path (private, secrets) or a credential-like string is refused and nothing is written. Edits are hash-anchored — patch replaces a paragraph by its content hash and is rejected if that content has moved, so concurrent edits never clobber.
@50firsttapes/core — schema system, validator, structural-health checks, query, ingest, governance gates, hash-anchored edits, OKF read/write.
@50firsttapes/cli — the tapes command: ingest · query · lint · write · govern.
@50firsttapes/mcp — a stdio or HTTP MCP server exposing the verbs to any client (Claude Code, Codex, claude.ai). The native tools — read (query/read/list/lint) and gated write (write/patch/govern) — plus wiki-mcp-compatible tools (session_bootstrap, search, read_note, list_folder, get_recent_logs) so it can stand in for the read-only wiki-mcp gateway.
@50firsttapes/remote — a git-remote adapter (Forgejo · GitHub · bare). 50 First Tapes depends on git, never on a forge.
pnpm install
pnpm build
node packages/cli/dist/index.js lint examples/sample-bundle
node packages/cli/dist/index.js query --kind note --bundle examples/sample-bundleRun the MCP server over a bundle. Stdio is the default (local clients like Claude Code):
TAPES_BUNDLE=/path/to/bundle node packages/mcp/dist/index.jsFor hosted use, serve over HTTP with Bearer auth (discovery is open; tool calls require a token):
TAPES_HTTP_PORT=8080 TAPES_MCP_TOKEN=sk-… TAPES_BUNDLE=/path/to/bundle node packages/mcp/dist/index.jsForgejo is the canonical remote; GitHub is a one-way push mirror. Contribute via Issues and PRs on GitHub — see CONTRIBUTING.md.
This server cannot be deployed
Git-backed platform for skills, tools, and context for AI agents
Governed, auditable knowledge your team curates for its AI assistants, self-hostable
Self-hostable shared brain for you and your AI agents — docs, flows, meetings, decisions, rationale
Open-source Obsidian for MDX - edit local docs with agent assistance
