Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
AppFlowy MCP
An MCP server and CLI for AppFlowy: 128 tools covering workspaces, spaces, pages, document content (incl. collab-level editing with snapshot/restore), Markdown import & export, PDF/ZIP export, databases & rows, search, members and invitations, sharing, comments, publishing, notifications, file upload, and AI chat & meeting transcription. Distributed as a single static Rust binary — the same appflowy binary is the MCP server (appflowy serve) and a full CLI.
Quick start
Install (details and more platforms: docs/installation.md):
# Linux / macOS
curl -fsSL https://raw.githubusercontent.com/weironz/appflowy_mcp/main/install.sh | sh
# Windows (PowerShell)
irm https://raw.githubusercontent.com/weironz/appflowy_mcp/main/install.ps1 | iexWire up an MCP client (docs/mcp-connect.md for Codex / Claude Desktop / Cursor, HTTP mode, and troubleshooting):
claude mcp add --scope user appflowy \
-e [email protected] \
-e APPFLOWY_PASSWORD=your-password \
-- appflowy serveOr drive it from a shell:
appflowy login
appflowy use my-workspace
appflowy export -o ./backup # whole workspace → Markdown tree (incremental on reruns)
appflowy search "quarterly report"Highlights
- Two personalities, one binary: MCP server (stdio by default, Streamable HTTP with
--http) and a scriptable CLI with workspace/PATH addressing. - Collab-faithful export: whole-workspace Markdown export with inline formatting decoded natively from yjs collabs (tables, math, nested lists), incremental on reruns; server-side PDF export and owner-only workspace ZIP backup.
- Full-fidelity import: Markdown files & directories (images uploaded in parallel, sibling order preserved), CSV → database, doc/docx/html/pdf → pages, Notion
.ziparchives, all as resumable server-side tasks. - Writes where it matters: pages, blocks, append (order-safe), collab-level document editing — replace a block's text, detach a block, snapshot and restore (CRDT-correct rebuild semantics) — plus database-row documents and row update/delete/restore.
- Agent-grade errors: typed errors keep HTTP status/code end-to-end; input validation returns
isError: truetool results (MCP SEP-1303) so the model can self-correct.
Documentation
| Where | What |
|---|---|
| docs/installation.md | Installers, manual/source install, uninstall, asset matrix |
| docs/mcp-connect.md | Client setup & troubleshooting (Codex, Claude, Cursor, HTTP) |
| docs/cli.md | CLI reference: auth chain, addressing, every command, cron/CI |
| docs/tools.md | The full 128-tool catalog |
| docs/agent-guide.md | How an AI client should use the tools (paste-ready block + recipes) |
| docs/development.md | Module map, design notes, adding a tool, release process |
| docs/ROADMAP.md | What's shipped and what's deliberately deferred |
| docs/README.md | Docs index |
Requirements & credentials
The server authenticates with environment variables (.env also works):
[email protected]
APPFLOWY_PASSWORD=your-password
APPFLOWY_BASE_URL=https://beta.appflowy.cloud # only for self-hostedThe CLI can use those, or persistent tokens from appflowy login (~/.config/appflowy-cli/credentials.json). Most write operations need a workspace plus a parent view id (workspace → space → page).
Notes
- The original Python implementation was removed in v0.8.0 (see tag
v0.8.0for those sources); PyPI packages are frozen atappflowy-mcp0.8.0 /appflowy-cli0.6.4. - Some endpoints are implemented from AppFlowy source routes that are not in the public OpenAPI document; newer commercial-cloud capabilities degrade with a clear 404 or a documented fallback on older self-hosted images.
- Logs always go to stderr (stdout stays the JSON-RPC channel).
License
Apache License 2.0 — Copyright 2026 weironz.