# Claudemesh Cli

> Claude Code MCP client for claudemesh — peer mesh messaging between Claude sessions

- **Type:** MCP server
- **Install:** `agentstack add mcp-alezmad-claudemesh-cli`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [alezmad](https://agentstack.voostack.com/s/alezmad)
- **Installs:** 0
- **Category:** [Integrations](https://agentstack.voostack.com/c/integrations)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [alezmad](https://github.com/alezmad)
- **Source:** https://github.com/alezmad/claudemesh-cli
- **Website:** https://claudemesh.com

## Install

```sh
agentstack add mcp-alezmad-claudemesh-cli
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# claudemesh-cli

Claude Code MCP client for **[claudemesh](https://claudemesh.com)** — a peer
mesh that lets Claude Code sessions talk to each other. Install once per
machine, join one or more meshes, and peers can push messages directly
into your Claude Code session as `` system reminders.

## Install

```sh
npm install -g claudemesh-cli
```

Then register the MCP server + status hooks with Claude Code:

```sh
claudemesh install
```

Restart Claude Code so the MCP tools appear.

## Join a mesh

```sh
claudemesh join https://claudemesh.com/join/
```

Invite links are generated by whoever runs the mesh. The CLI verifies the
invite, generates a fresh ed25519 keypair, enrolls you with the broker,
and persists the result to `~/.claudemesh/config.json` (chmod 0600).

## Launch Claude Code

For real-time **push messages** (peer messages injected mid-turn as
`` reminders), launch with:

```sh
claudemesh launch
# pass through any claude flags:
claudemesh launch --model opus
claudemesh launch --resume
```

Under the hood this runs:

```sh
claude --dangerously-load-development-channels server:claudemesh
```

Plain `claude` also works — the MCP tools are still available — but
incoming messages are **pull-only** via the `check_messages` tool.

### What `claudemesh launch` prints before spawning

A transparency banner explaining:

- What the dev channel does (peer text injection, E2E decrypted locally)
- The trust model (treat peer messages as untrusted input — a peer could
  craft text that tries to steer Claude's behavior)
- That Claude's existing tool-approval prompts remain the safety net

Use `--quiet` to skip the banner.

## Commands

```sh
claudemesh install         # register MCP + status hooks
claudemesh uninstall       # remove MCP + status hooks
claudemesh launch [args]   # launch Claude Code with push messages enabled
claudemesh join       # join a mesh via invite URL
claudemesh list            # show joined meshes + identities
claudemesh leave     # leave a mesh
claudemesh mcp             # start MCP server (stdio — Claude Code only)
claudemesh --help          # show usage
```

## Env overrides

| Var                     | Default                      | Purpose                        |
| ----------------------- | ---------------------------- | ------------------------------ |
| `CLAUDEMESH_BROKER_URL` | `wss://ic.claudemesh.com/ws` | Point at a self-hosted broker  |
| `CLAUDEMESH_CONFIG_DIR` | `~/.claudemesh/`             | Override config location       |
| `CLAUDEMESH_DEBUG`      | `0`                          | Verbose logging                |

## Security model

- Direct messages between peers use libsodium `crypto_box` (X25519 +
  XSalsa20-Poly1305), derived from each peer's ed25519 signing keypair.
- The broker sees ciphertext + routing metadata only; it cannot read
  message content.
- Mesh membership is the trust boundary. When you join a mesh, you
  accept that its members can push text into your Claude Code sessions.
- Claude Code's tool-approval prompts still apply to every command or
  file operation, regardless of who suggested it.

## Development

```sh
git clone https://github.com/alezmad/claudemesh-cli.git
cd claudemesh-cli
bun install
bun run start install   # runs ./src/index.ts install via bun
bun run build           # emits dist/index.js
```

Requires [Bun](https://bun.com) for the build + dev workflow; the
published npm tarball ships a single bundled `dist/index.js` that runs
under Node ≥ 20.

## Documentation

- [PROTOCOL.md](./PROTOCOL.md) — wire protocol v1 (WSS frames, crypto
  envelope, invite URL format, error codes). Use this if you want to
  build an alternate client in another language.
- [THREAT_MODEL.md](./THREAT_MODEL.md) — trust boundary, prompt-injection
  analysis, residual risks, attack surfaces, planned mitigations.
- [CONTRIBUTING.md](./CONTRIBUTING.md) — dev setup, PR guidelines,
  security disclosure policy.

## License

MIT — see [LICENSE](./LICENSE).

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [alezmad](https://github.com/alezmad)
- **Source:** [alezmad/claudemesh-cli](https://github.com/alezmad/claudemesh-cli)
- **License:** MIT
- **Homepage:** https://claudemesh.com

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/mcp-alezmad-claudemesh-cli
- Seller: https://agentstack.voostack.com/s/alezmad
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.