Docs

Appearance

Use Your MCP Server ​

If your system already speaks MCP, your agent can use its tools on calls: check stock, reserve a slot, read an order. You add the server and tick the tools the agent may call. Rymi connects to it at the start of each call and offers the agent only those tools, under the server's own names.

In Studio this is Tools → Developer → MCP server. Through the API it's an mcp_server entry in the agent's tools array, one per server.

Add one ​

  1. Open Tools → MCP server and click Add MCP server. Give it a Name, such as Inventory. Only Studio shows it.
  2. Enter the Server URL: the streamable-HTTP endpoint, such as https://mcp.example.com/mcp. It must be a public https URL.
  3. If the server needs a token, add an Authorization header and pick Use secret → New secret…. The secret is pinned to the server's host.
  4. Click Find tools. Rymi lists what the server offers.
  5. Tick the tools the agent may call, then save. The live agent uses them from its next call.

Tick only what the agent needs. Each tool can change data on your server, and the agent decides when to call it from its instructions and the tool's own description.

What happens on a call ​

  • Rymi connects to each enabled server when the call starts and waits up to 3 seconds. A server that is down or slow is skipped; the call goes on without its tools.
  • A server whose header references a secret you haven't saved is skipped too. Rymi never sends a blank key.
  • A server tool named like a built-in tool, such as end_call, is skipped. If two servers offer the same tool name, the first server's tool wins.
  • Turn a server off with its switch to keep its settings without using it.

Through the API or SDKs ​

See MCP Server Bindings for the binding and Discover MCP Tools for listing a server's tools. Read the agent first and send back its whole tools array, so you don't drop its other tools. The Rymi MCP's add_agent_tool edits API tools only; add MCP servers in Studio or with an agent update.