One key. Every agent.
One org key. Every agent. Manage your whole Hermes fleet from Cursor, Claude, or Codex. Lifecycle, secrets, and ops without SSH.
One org key. Every agent in your fleet.
List them. Aim at one. Operate from Cursor, Claude, or Codex. SEAOTTER Connect is one admin layer for every always-on Hermes in your workspace.
One key. The whole org.
Every SEAOTTER workspace is an organization. Your MCP access key (so_…) is scoped to that org, not to a single agent.
One key. Three moves:
- List the fleet (
list_agents) - Aim at the right one (
agent_id) - Operate: lifecycle, secrets, health, chat
No second key when you add a client. No credential swap when you switch engagements. Create a key once at seaotter.dev/connect. Install MCP once. Work across the fleet.
Full setup: Connect from chat.
List first. Then aim.
Step one: list_agents. Names, IDs, status, regions, dashboard URLs.
Step two: agent_id on every tool that touches a specific agent: chat (talk_to_hermes_agent), lifecycle (start_agent, stop_agent, restart_agent, delete_agent), secrets (upsert_secret, delete_secret), backups (list_agent_backups, restore_agent), runtime checks (hermes_status, hermes_logs).
One key. Many agents. Always explicit about which one.
Paste this into chat after Connect is installed:
Use the SEAOTTER MCP: call list_agents first. If none exist, onboard a Hermes agent; if several exist, ask which agent_id to use, then talk_to_hermes_agent with that agent_id.
A morning on the fleet
You open Cursor. Same so_… key as yesterday.
list_agents returns four running agents: Acme, Brightline, Cedar, Delta. You stop_agent on Cedar, upsert_secret on Brightline, pull hermes_logs on Acme, then talk_to_hermes_agent to rerun the job.
Four clients. One connection. No credential swap.
What org admins can do from chat
MCP is the control plane in your IDE. Use it when you are already coding or coordinating. The Hermes dashboard is still there for browser-only setup.
| From MCP | |
|---|---|
| See | list_agents, get_agent |
| Provision | create_agent, create_agent_and_wait |
| Pause | stop_agent |
| Resume | start_agent, restart_agent |
| End | delete_agent (confirm: true) |
| Secrets | upsert_secret |
| Recover | list_agent_backups, restore_agent |
| Diagnose | hermes_status, hermes_logs, hermes_doctor |
| Work | talk_to_hermes_agent with the right agent_id |
| Billing | get_subscription, list_plans |
Delete, restore, and credential rotation require confirm: true. Model OAuth, channel pairing, and full hermes setup hand you a dashboard URL when chat is not the right surface.
Built for agencies
If you package one isolated Hermes per client, fleet MCP is the ops layer beside the isolation story: separate namespaces, separate secrets, one admin key that sees the whole roster.
Start with one agent on a 7-day trial, no credit card. Add agents at $99 per month each when you are ready to scale. Book Custom for SSO or negotiated quotas.
See pricing.
Security
The so_… key is org admin access. Treat it like a password.
- One key per MCP client (Cursor, Claude, Codex), not per agent.
- Revoke from Settings when someone leaves the team.
- Never paste Hermes dashboard passwords into MCP config.
- Put integration keys in Secrets via
upsert_secret. How Secrets works.
Details: MCP security.
The fleet, where you work
Provision on the plane. Connect once. Run the fleet where you work.