Hermes Bridge

Hermes Bridge

Enables MCP agents to delegate tasks to a local Hermes Agent for terminal, file, browser, and coding operations, and schedule recurring jobs.

Category
Visit Server

README

Odin → Hermes Bridge

Give Odin (or any MCP-compatible agent) a way to hand off tasks it can't do itself to a local Hermes Agent install — full terminal, file, browser, and coding tool access — and get a real answer back. Also supports scheduling recurring jobs (morning routines, daily reports) that Hermes runs on its own.

No webhooks, no servers, no Telegram relay for tasks. Just one Python file that Odin talks to directly on your own machine.

Fastest setup: let Hermes install itself

If you already have Hermes running and talk to it on Telegram, just message it:

install this: https://github.com/182moon/odin-hermes-bridge

Hermes will download the bridge, install its one dependency, detect its own install paths, and hand you back a ready-to-paste MCP config with your actual Telegram chat ID already filled in — no placeholders to edit. You just paste that block into Odin's MCP Servers → + Add panel. That's the only manual step left (Hermes can't click Odin's UI for you).

Prefer to do it yourself in a terminal instead of asking Hermes? Run:

curl -fsSL https://raw.githubusercontent.com/182moon/odin-hermes-bridge/main/install.sh | bash -s -- YOUR_TELEGRAM_CHAT_ID

(Omit the chat ID argument if you don't know it yet — the printed config will have a placeholder you can fill in by hand.)

The rest of this README covers the same setup done fully manually, plus how everything works under the hood.

What you get

Once set up, Odin gains four new tools:

Tool What it does
run_hermes_task Hand off a one-off task Odin can't do itself — Hermes runs it now and returns the result
schedule_hermes_task Set up a RECURRING job (a morning routine, daily report, etc.) that Hermes runs on its own schedule, independent of Odin
list_scheduled_tasks See what's currently scheduled
remove_scheduled_task Cancel a scheduled job

run_hermes_task runs a full agent loop for a single task and returns the answer — use it for anything "do this now." schedule_hermes_task sets up a recurring job inside the user's own Hermes install using Hermes's built-in cron scheduler — use it for "do this every morning / every day / every Monday," etc. Scheduled jobs keep running even when Odin isn't open, as long as Hermes's own gateway/scheduler process is running on the user's machine (check with hermes cron status).

Important — where scheduled results go: Odin only creates the job; it has no way to receive results back later, since the job runs completely independently. schedule_hermes_task delivers results to a Telegram chat (the chat ID you already use with Hermes), not back into Odin. If you don't pass a telegram_chat_id (or set HERMES_TELEGRAM_CHAT_ID as a default — see Configuration knobs below), the job is created with deliver=local and its output only shows up in list_scheduled_tasks / hermes cron list — nobody gets notified. For an actual "morning briefing that lands on your phone," you need Hermes's Telegram gateway already connected (hermes gateway setup) and pass that chat's ID.

Requirements

  • macOS, Linux, or WSL (same as Hermes itself)
  • Python 3.9+
  • Hermes Agent installed and configured with a working model (you'll use your own OpenRouter/ Anthropic/etc. key, same as Odin)

Setup (5 minutes)

1. Install Hermes, if you haven't already

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
hermes setup

Confirm it works:

hermes chat -q "say hello"

2. Download this bridge folder

Save hermes_task_server.py and requirements.txt from this folder somewhere on your machine, e.g. ~/hermes-mcp-bridge/.

3. Install the one dependency

pip3 install -r ~/hermes-mcp-bridge/requirements.txt

(This installs the mcp Python package — the bridge script needs it to speak the MCP protocol. It's separate from Hermes itself.)

4. Find your hermes path

which hermes

Copy that path — you'll need it in the next step if hermes isn't already on Odin's PATH.

5. Add Hermes as an MCP server in Odin

Open Odin → MCP Servers+ Add → paste config, editing the paths to match your machine:

{
  "mcpServers": {
    "hermes": {
      "command": "python3",
      "args": ["/Users/yourname/hermes-mcp-bridge/hermes_task_server.py"],
      "env": {
        "HERMES_BIN": "/Users/yourname/.local/bin/hermes"
      }
    }
  }
}

Save. Odin should now show run_hermes_task as an available tool.

6. Test it

In Odin, ask it something you know it can't do on its own — e.g. "use Hermes to list the files in my Downloads folder" — and confirm it calls the tool and comes back with a real answer.

Then test scheduling — ask Odin something like "set up a Hermes job that runs every morning at 7am and gives me a weather + calendar briefing." Confirm it shows up with hermes cron list in a terminal, or ask Odin to list your scheduled tasks.

7. Keep the scheduler running

Recurring jobs created with schedule_hermes_task only fire while Hermes's scheduler is actually running in the background — not just when Odin happens to be open. Check status with:

hermes cron status

If it's not running, either start it in the foreground for testing:

hermes gateway run

(stays alive only while that terminal window is open — good for a quick test, not for production)

or install it as a real background service that survives reboots and terminal closures:

hermes gateway install
hermes gateway start
hermes gateway status   # confirm it's running

How it works

One-off tasks (run_hermes_task):

  1. Odin calls run_hermes_task(task="...")
  2. The bridge script runs hermes chat -q "<task>" as a subprocess on your machine
  3. Hermes runs its full agent loop (its own tools, memory, skills — all of it) until the task is done
  4. The final text answer is returned to Odin as the tool result

Recurring tasks (schedule_hermes_task):

  1. Odin calls schedule_hermes_task(schedule="every day at 7am", task="...")
  2. The bridge runs hermes cron create under the hood, registering a real job in the user's own Hermes install
  3. Hermes's own scheduler fires the job on schedule — with or without Odin open — as long as the Hermes gateway/scheduler is running
  4. list_scheduled_tasks / remove_scheduled_task wrap hermes cron list / hermes cron remove so Odin can manage jobs it created

Every call is independent — pass everything the calling agent knows about the task into the task text, since Hermes can't see Odin's conversation history or previous scheduled runs.

Configuration knobs

Env var Purpose Default
HERMES_BIN Full path to the hermes executable first match on PATH
HERMES_TASK_TIMEOUT Max seconds to wait for a task 280
HERMES_TELEGRAM_CHAT_ID Default Telegram chat ID for schedule_hermes_task results, so Odin doesn't need to ask for it every time none — must be passed per call if unset

Find your Telegram chat ID by messaging your Hermes bot once, then checking hermes gateway status or your Hermes logs for the chat ID, or by asking Hermes itself "what's this chat's ID" in a Telegram conversation with it.

Set these in the "env" block of the MCP config shown above.

Troubleshooting

"No module named mcp" Run pip3 install mcp using the exact python3 your MCP config's "command" points to. If you have multiple Python installs, use the full path in both the install command and the config, e.g. /usr/bin/python3 -m pip install mcp.

"could not find the 'hermes' executable" Set HERMES_BIN in the MCP config's "env" block to the output of which hermes.

Tool call times out on long tasks Raise HERMES_TASK_TIMEOUT in the "env" block (seconds), or ask Odin to pass a higher timeout_seconds per call.

Odin doesn't see the tool after adding the server Restart Odin, or check its MCP server list for a connection error — usually a wrong path in "args" or "env".

Security note

This bridge runs Hermes with whatever tool access and permissions your Hermes install already has configured (terminal commands, file access, etc.). Anything Odin delegates through this tool runs with those same permissions on your machine. Only connect this to agents you trust.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured
E2B

E2B

Using MCP to run code via e2b.

Official
Featured