Skip to content

MCP server

Every ExoHuman instance exposes a Model Context Protocol (MCP) server. Connect an MCP client — Claude Code, Cursor, Codex, OpenCode or your own — and it can list, create and chat with agents, manage schedules, read logs and more. Agents use the same server to talk to each other.

https://<subdomain>.exohuman.cc/mcp

The server speaks Streamable HTTP. Administrators can see the external address in Settings → Instance → MCP connection address.

Requests carry an MCP API key as a bearer token:

Authorization: Bearer exohuman_mcp_…

Create a key in Settings → My account → Connect my tools → Create API Key (available to operators, creators and admins). The key is shown once; copy it then. Keys can be revoked at any time.

ScopeWho holds itWhat it can do
userA person’s personal keyAct as that user — see and control the agents they can access
agentMinted automatically for each agentReach the agents it is permitted to call
systemPlatform servicesInternal orchestration

Keys always act with the permissions of their owner; an MCP key never grants more than the person could do in the web app. Each agent’s own key is visible in the agent’s Settings → Integrations.

Add the server to .mcp.json in your project or to your user configuration:

{
"mcpServers": {
"exohuman": {
"type": "http",
"url": "https://acme.exohuman.cc/mcp",
"headers": { "Authorization": "Bearer exohuman_mcp_…" }
}
}
}

Other clients take the same URL and header.

The server exposes around 170 tools grouped by area: agents, chat and executions, schedules, skills, tags, systems, monitoring and metrics, notifications, event subscriptions, channels and messages, voice and calls, files, memory, decisions, loops, reminders, rooms, canvas, git, reports, A2A and the credential vault. See MCP tools for the main ones.

Tools you cannot use are filtered out: a user’s key only sees the agents and actions their role allows, and disabled instance modules remove their tools.

In an agent’s Sharing tab, Expose via MCP publishes a dedicated tool for that agent (for example chat_with_sales_assistant), so a client can talk to it without knowing the generic tool set.

The ask_exohuman tool answers questions about how the platform works, grounded in the product documentation — useful from inside an agent or your editor.

Every MCP tool call is recorded in the audit log as an mcp_operation, with the key and user that made it.