Hermes Bridge
Enables MCP agents to delegate tasks to a local Hermes Agent for terminal, file, browser, and coding operations, and schedule recurring jobs.
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):
- Odin calls
run_hermes_task(task="...") - The bridge script runs
hermes chat -q "<task>"as a subprocess on your machine - Hermes runs its full agent loop (its own tools, memory, skills — all of it) until the task is done
- The final text answer is returned to Odin as the tool result
Recurring tasks (schedule_hermes_task):
- Odin calls
schedule_hermes_task(schedule="every day at 7am", task="...") - The bridge runs
hermes cron createunder the hood, registering a real job in the user's own Hermes install - Hermes's own scheduler fires the job on schedule — with or without Odin open — as long as the Hermes gateway/scheduler is running
list_scheduled_tasks/remove_scheduled_taskwraphermes cron list/hermes cron removeso 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.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。