Skip to main content
Glama
sc28249782

Hermes MCP Bridge

by sc28249782
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

Hermes MCP Bridge

Securely connect ChatGPT to a local Hermes Agent and Codex CLI running in WSL2.

Tests License

Hermes MCP Bridge is a local, stdio-based MCP bridge for controlled work on a user's own WSL2 machine. It exposes Hermes task operations and a permission-gated Codex execution layer through an OpenAI Secure MCP Tunnel.

Read Project origin and Project history.

What it provides

Area

Capabilities

Hermes

Health checks, model catalog, task submit/status/result/cancel, recent tasks, usage summary and export

Codex / WSL2

Health, submit, status, paginated result, cancel, and recent jobs

Security

Workspace allowlist, symlink-escape protection, explicit local approval for write jobs, fixed argv invocation, timeout and prompt-size limits

The v0.4.0 design exposes 16 MCP tools: 10 Hermes tools and 6 Codex tools.

Related MCP server: chatgpt-deepseek-bridge

Security model

Codex jobs are deliberately restricted:

  • Allowed sandboxes: read-only and workspace-write only.

  • danger-full-access is rejected.

  • Every workspace must be explicitly listed in codex.allowed_workspaces.

  • A workspace-write job remains pending until a person approves it locally with codex-approve.

  • An MCP client cannot approve its own job.

  • Codex is started without a shell, using fixed argument handling.

Never commit bridge-config.json, state/, logs, virtual environments, runtime API keys, or Hermes secrets.

Architecture

ChatGPT
  → OpenAI Secure MCP Tunnel
  → bridge.sh (stdio MCP)
  → Hermes API on 127.0.0.1:8642
  → Codex CLI / approved WSL2 workspaces

Quick start

Run these commands as your normal WSL2 user; do not use sudo.

unzip hermes-mcp-bridge-v0.4.0.zip
cd hermes-mcp-bridge-v0.4.0
bash install.sh

./bridge.sh doctor
./bridge.sh codex-doctor

Then configure the allowed project roots in bridge-config.json:

{
  "api_url": "http://127.0.0.1:8642",
  "hermes_env": "/home/USER/.hermes/.env",
  "hermes_config": "/home/USER/.hermes/config.yaml",
  "codex": {
    "allowed_workspaces": [
      "/home/USER/projects/example"
    ]
  }
}

Start the Secure MCP Tunnel after completing its one-time configuration:

bash tunnel.sh run

See the Thai documentation for the full setup and live acceptance procedure:

Status

The automated v0.4.0 validation covers Hermes regression, Codex permission/process behavior, MCP discovery, and tunnel security. Perform the live WSL2 acceptance checklist before treating Codex execution as production-ready.

License

Licensed under Apache License 2.0.

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers