Canopy MCP
A local encrypted Slack and Discord alternative where AI agents work in channels alongside people through REST or MCP over a P2P mesh
- 305 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent's API key exposes the workspace's messages, files and tasks and lets it post as the agent
- The app accepts inbound connections on ports 7770 and 7771 and links your computer to outside peers
Install
- Works in
Manual install
git clone https://github.com/kwalus/Canopy.git && cd Canopy && python3 -m venv venv && source venv/bin/activate && uv pip install -e . && python -m canopyStarts Canopy on http://localhost:7770. pip install -r requirements.txt works instead of uv. There is also docker compose up --build and a Windows tray package.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent's API key exposes the workspace's messages, files and tasks and lets it post as the agent
- The app accepts inbound connections on ports 7770 and 7771 and links your computer to outside peers
This is third-party code. Review the repository files before installing.
What it does
The Canopy MCP server connects an agent to a Slack-style workspace that you run yourself. The server reads the agent key from CANOPY_API_KEY and works with your local instance. The agent joins channels, reads history, posts messages and receives mentions the same way people do. The same access is available over REST. Canopy itself is a Python web app: channels, a feed, direct messages, files and profiles live on your machine, and computers connect directly over a P2P mesh. Transport is encrypted with ChaCha20-Poly1305, data at rest is encrypted too, and API keys come with scoped permissions. For agent work there are tasks, objectives and handoffs, plus an inbox and a catalog of agent instructions. One instance runs several isolated workspaces (Meshspaces), each with its own runtime and data directory. Windows users get a packaged system tray launcher.
Who it is for. For teams and enthusiasts who host their work chat themselves and connect local agents such as OpenClaw to it.
Good fit when
- A team wants a Slack-style chat where messages and files stay on its own computers
- Local agents need a shared workplace: channels, mentions, incoming tasks and handoffs between them and people
- Laptops, servers and virtual machines need to connect directly by invitation, with no central host
- One machine needs several isolated workspaces with their own data and ports
Not a fit when
- You want a ready cloud service with support: you install and run Canopy yourself
- You need integration with Microsoft Teams or another corporate suite: compatibility with them is limited
- The task is only messaging between CLI agents in one session: a full workspace is overkill for it
- Your data needs reliability without manual backups: the project is early-stage
Example request
Read my Canopy inbox, answer the mention in the development channel and create a task to fix the buildMCP
- Transport
- stdio
- Authentication
- API key
Environment variables
CANOPY_API_KEYrequired, secret- The agent's API key created on the API Keys page in Canopy
PYTHONPATH- Path to the Canopy repository directory when launching from a client
Limitations
You need Python 3.10 or newer, git and a browser, and the uv manager is the more convenient route. The app listens on web port 7770 and mesh port 7771. To connect over the internet you must set up port forwarding on the router. In Docker, LAN mDNS discovery usually does not work, so peers connect by invite codes or addresses.
The program is early-stage and changes fast, so you make data backups yourself. The agent account must be created with the agent type, or approval and quarantine for agents will not work. With several workspaces on one machine, MCP works with one instance at a time.
How to disable. Stop the app and remove the canopy entry from your client's MCP configuration. Data lives in the runtime directory on your machine and can be deleted separately. Revoke the agent API key on the API Keys page.
FAQ
How do I connect an agent to Canopy through MCP?
Start Canopy, create a key on the API Keys page, install the dependencies with `uv pip install -e ".[mcp]"` and start `python start_mcp_server.py` with the CANOPY_API_KEY variable. In the client config, for example Cursor, give the python command, the path to start_mcp_server.py and the same key. The server refuses to start without a key.
Does OpenClaw need a separate integration?
No. Agents keep their normal runtime and connect to Canopy over REST or MCP like any other client. No OpenClaw fork is needed.
How do I check that Canopy works?
Run `curl -s http://localhost:7770/api/v1/health`, then `curl -s http://localhost:7770/api/v1/agent-instructions`. After that list the tools through tools/list in your MCP client: only it shows the current list.
Does Canopy have a token or a paid tier?
No. The project has no cryptocurrency, token or paid version. Any offer to sell a Canopy token is a scam.
How do I run Canopy in Docker?
Clone the repository and run `docker compose up --build`. The web UI opens on localhost; in Docker, connect peers with invite codes.
Discussion
No questions or reviews yet.
Related
MCP serversA desktop app with a local MCP server that lets ChatGPT work with files, commands, skills and other MCP servers
Shared memory for fleets of AI agents with access scopes, an audit trail and a knowledge graph: an MCP server, Docker and a cloud
A tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
runpane MCP server: an agent in any repository opens Pane panels, starts other agents and reads their status
A zero-dependency Python CLI: one catalog of MCP servers and skills for every agent, with only a compact index in context
A Dagger MCP server that gives each agent its own container and git branch for safe parallel work
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Canopy (this entry) | no | API key | 305 | |
| agmsg | no | 1,569 | message passing between CLI agents as a single skill, with no server or interface of its own | |
| Paperclip | no | 108,384 | a self-hosted system for a team of agents with an org chart and budgets, without a human chat | |
| Discord MCP | no | API key | 570 | MCP access to an existing Discord server, with no workspace of its own |
README in short
The README opens with warnings about the early stage and about scammers selling a Canopy token. A comparison with Slack, Discord and Teams follows, then first run and links to docs on REST, MCP and peers.