Agent Mesh CLI
Agent Mesh links agents, containers and laptops into a private network where nodes publish tools and models over MCP
- 1,186 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- A node publishes local tools and models to the network, and the network policy decides which nodes may call them
- The install script is downloaded and run with `curl | bash`, so review it first
- Node tokens and credentials sit on disk, so keep a node on a trusted machine only
Install
- Works in
Manual install
curl -sL https://google.github.io/agentmesh/install.sh | bashInstalls `agentmesh-node`, `mcp-client` and other programs. Alternative: `go install github.com/google/agentmesh/cmd/agentmesh-node@latest`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- A node publishes local tools and models to the network, and the network policy decides which nodes may call them
- The install script is downloaded and run with `curl | bash`, so review it first
- Node tokens and credentials sit on disk, so keep a node on a trusted machine only
This is third-party code. Review the repository files before installing.
What it does
Agent Mesh is a private network for AI agents from Google. The `agentmesh-node` program runs next to an agent and does three things: publishes your tools and models to the network, finds and calls what other nodes publish, and presents a verifiable identity. Nodes connect directly when they can and through relays when they cannot, so the network works behind NAT on laptops, in containers, in clusters and on phones. For the agent the node starts a local MCP server on `127.0.0.1:8080` and an OpenAI-compatible endpoint that routes model requests to the node hosting the model. The `agentmesh` skill in `agents/skills` teaches the agent to look for a capability locally first and to reach into the network only when the task cannot be done without it. Install the skill with `agentmesh-node skill install`. Nothing is reachable by default: a node exposes no services until its configuration says so, and calls are granted by policy on service names rather than addresses. Identity comes from your OpenID Connect provider, and the control plane turns a token into a short-lived, offline-verifiable credential. Besides the node, the project ships `agentmesh-control-plane`, `agentmesh-router` and `agentmesh-one` for laptops and small deployments.
Who it is for. For developers who need agents on different machines to find each other's tools and models without open ports and shared keys.
Good fit when
- A model is hosted on one computer while the agent runs on another: the node's OpenAI-compatible address routes the request to the right machine
- An agent in a container or behind NAT must call MCP tools published on a laptop or in a cluster
- You need an access policy on service names with identity from your OpenID Connect provider
- You want to try a mesh quickly: `agentmesh-node join` against the shared testnet or `agentmesh-one` on your laptop
Not a fit when
- All tools and models sit on one machine: local MCP servers are enough without a network
- You want a ready cloud service with support: the project is self-run and self-maintained
- You have no OpenID Connect provider and no wish to manage enrollment tokens for nodes
- You only want to publish an MCP server to the internet: Agent Mesh builds a closed network with no public address
Example request
Check through Agent Mesh which services are available on the network and call the model a neighboring node publishedLimitations
The shared `bananas.sam-mesh.dev` testnet is for developers and promises no uptime, so run your own control plane for steady work. Sandboxed agents and the mobile app are still in preview and keep changing. Before nodes exchange calls, you describe services in the node configuration and rules in the network policy. Installation uses a `curl` script or `go install`.
How to disable. Stop the node and remove the skill by deleting the `agentmesh` folder from the agent's skills directory. The installer puts binaries in the directory its output names, so delete them manually.
FAQ
How do I install Agent Mesh?
Run `curl -sL https://google.github.io/agentmesh/install.sh | bash`, which installs `agentmesh-node`, `mcp-client` and other programs. `go install github.com/google/agentmesh/cmd/agentmesh-node@latest` works instead of the script.
How do I connect an agent to Agent Mesh?
Join with `agentmesh-node join <address>`, start the node with `agentmesh-node run --daemonize` and install the skill with `agentmesh-node skill install`. The node starts a local MCP server on 127.0.0.1:8080 that you can add to the agent's client.
Do I need my own server to try it?
No. A shared `bananas.sam-mesh.dev` testnet exists for trials with no uptime promise. `agentmesh-one` runs a control plane on your laptop, and in Codespaces `make testnet` is enough.
Who issues identity to nodes?
Your OpenID Connect provider or a bootstrap token. The control plane turns it into a short-lived credential bound to the node's key, which other nodes verify without contacting the server.
Can I call models through Agent Mesh?
Yes. The node offers an OpenAI-compatible address that routes requests to the node hosting the model. The `agentmesh` skill describes the order for calling inference and inspecting network services.
Discussion
No questions or reviews yet.
Related
CLI toolsAn open coding agent for the terminal, browser and desktop that works with models from OpenAI and Claude to Ollama
A hub and the anet CLI join Claude Code, Codex, Grok and OpenCode from different machines into a network driven from one window.
AgentBridge by Graphene Lab
AgentBridge Graphene Lab
The standalone AgentBridge server exposes over MCP one office agent with Word, Excel, email and a task scheduler
A plugin that gives an agent the IntelliNode skill and MCP server: ask any model, search, review code, tests, SQL, images and speech
A personal AI agent for the terminal with shell, Python, web and local models, one of the earliest agent CLIs
A desktop app with a local MCP server that lets ChatGPT work with files, commands, skills and other MCP servers
README in short
Agent Mesh consists of a control plane, a router and a node that runs next to an agent. The docs cover a quick start, concepts, guides for Kubernetes and Cloud Run and a reference for flags and policies.