How do I install Hostmesh?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Hostmesh need an API key?

No required environment variable is listed on this record.

What transport does Hostmesh use?

Hostmesh runs locally over stdio.

When was Hostmesh last checked?

Last checked 2026-10-28. Local-first control plane that gives MCP-compatible AI agents safe, structured, and auditable access to Linux servers over SSH.

Hostmesh MCP सर्वर

Power your agent

यह सर्वर जोड़ें

Install from the publisher. Not a PlugMCP endpoint.

PlugMCP पर चलाएँ

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

PlugMCP पर चलाएँ

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

लॉन्च

Server facts

Local

Transport: stdio
License: MIT
Last checked: 2026-10-28

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

HostMesh

Give AI agents safe, structured access to your Linux servers.

HostMesh is a local-first control plane for MCP-compatible clients such as Cursor, Claude Code, and Codex. The Portal, policy engine, identities, audit log, and SQLite database stay on your machine; remote Linux hosts run a deterministic agent reached through SSH.

Quick install

Install the Local Control Plane on macOS or Linux (amd64 or arm64):

curl -fsSL https://github.com/sotarak/hostmesh/releases/latest/download/install-hostmesh.sh | bash

The installer downloads the release archive, verifies its SHA-256 checksum, installs hostmesh into ~/.local/bin, and starts a per-user service. It uses LaunchAgent on macOS and a systemd user service on Linux. No root daemon or shell-profile edit is required.

The Portal opens at http://127.0.0.1:47821. To inspect the script before running it, download install-hostmesh.sh from the release page first.

To build instead:

git clone https://github.com/sotarak/hostmesh.git
cd hostmesh
make build
./bin/hostmesh start

See Installation for upgrades, service locations, manual builds, and uninstall options.

Add your first server

Open the Portal and create a workspace.
Go to Servers → Add server.
Enter the SSH host, port, user, and labels.
Copy the generated command and the one-time enrollment token.
Run the command on the Ubuntu server and enter the token when prompted.

curl -fsSL https://github.com/sotarak/hostmesh/releases/latest/download/install-agent.sh | sudo bash

The token is read from /dev/tty, so it does not appear in shell history or process arguments. It is short-lived, single-use, and bound to one workspace and server. The agent listens only on 127.0.0.1:7443; HostMesh reaches it through an SSH tunnel.

Core features

Local Portal, API, SQLite state, controller identity, and one stdio MCP server.
Workspace-scoped server registry and label selectors.
SSH-first enrollment with permanent Ed25519 controller/agent trust.
Signed, target-bound controller requests and agent responses.
Semantic system, process, file, service, and network inspection.
Reusable custom policies with managed templates and server attachments.
Deny-by-default evaluation with DENY > APPROVAL > ALLOW.
Agent-side hard denies, bounded execution/output, and two-phase audit events.
Local server removal separated from remote agent uninstall.

Current implementation status and future work live in Project status and roadmap, not in this README.

How it works

Cursor / Claude Code / Codex
             │
             │ MCP (stdio)
             ▼
┌─────────────────────────────┐
│ HostMesh Local Control Plane│
│ Portal · Workspaces · Policy│
│ Connections · Audit · SQLite│
└──────────────┬──────────────┘
               │ SSH tunnel
               ▼
┌─────────────────────────────┐
│ Linux server                │
│ hostmesh-agent              │
│ 127.0.0.1:7443              │
└─────────────────────────────┘

Every remote operation follows the same path:

MCP → workspace target resolution → persisted policy → audit start
→ SSH transport → signed agent request → agent validation
→ signed response → audit completion

Policies

The Portal includes reusable policy templates:

Default read-only diagnostics for the normal inspection workflow.
Log files reader for file.read beneath /var/log/** only.
System inspector for system, process, service, and network metadata without file access.

Templates are cloned into editable custom policies and attached to servers. Use Replace current policies when reducing a server to least privilege; adding policies unions their rules. No matching rule means deny.

MCP setup

Ensure hostmesh is on the MCP client's PATH, then configure one local MCP server:

{
  "mcpServers": {
    "hostmesh": {
      "command": "hostmesh",
      "args": ["mcp"]
    }
  }
}

Choose the default workspace with hostmesh workspace use <workspace-id>. Tools target a server by name/ID or a bounded label selector; HostMesh never creates one MCP namespace per server. See MCP.

Uninstall

Remove a remote Ubuntu agent while preserving its identity/state:

curl -fsSL https://github.com/sotarak/hostmesh/releases/latest/download/uninstall-agent.sh | sudo bash

Remove the local controller while preserving local data:

curl -fsSL https://github.com/sotarak/hostmesh/releases/latest/download/uninstall-hostmesh.sh | bash

Both uninstallers support --purge; destructive purge requires an interactive confirmation or explicit --yes. Removing a server in the Portal only removes its local registry data—it never SSHes into the host or uninstalls the agent.

Security

HostMesh is localhost-only and local-first, but it is not yet presented as production-grade security software. Important controls include workspace isolation, one-time enrollment, permanent identities, signed target-bound protocol messages, policy resource constraints, agent hard denies, checksum-verified artifacts, and bounded audit-safe operations.

Review the threat model and current limitations in Security. Report vulnerabilities through SECURITY.md, not a public issue.

Documentation

Contributing

Focused issues and pull requests are welcome. Read CONTRIBUTING.md and AGENTS.md before changing protocol, policy, enrollment, transport, or other trust boundaries.

License

MIT