MCP tools¶
Status: Current | Last reviewed: 2026-06-30
The Model Context Protocol (MCP) is an open standard for connecting AI models to external tools and data sources. An MCP tool points TrueServe at an MCP server; the tools that server exposes become callable by your agents and workflows.
How MCP works¶
MCP is client-server:
- An MCP server exposes a set of tools, each with a name, description, and parameter schema.
- You configure a connection to that server as an MCP tool in TrueServe.
- During a conversation, the LLM sees the server's tools as callable functions.
- When the LLM calls one, TrueServe forwards the request to the MCP server, gets the result, and feeds it back to the LLM.
Configuration¶
When you add an MCP tool, you set:
| Field | Required | Notes |
|---|---|---|
| MCP Server URL | Yes | The server endpoint, e.g. https://api.example.com/mcp. |
| Transport | Yes | Streamable HTTP (default) or Server-Sent Events (SSE). Pick what your server speaks. |
| Authentication | No | None, API Key, Bearer, Basic, Integration, or OAuth 2.0 — see Tools overview. |
| Custom headers | No | Static, non-secret metadata headers sent during discovery and runtime calls, such as tenant or server-selection headers. Use Authentication for API keys, bearer tokens, cookies, and other secrets. |
| Tool filters | No | Restrict which of the server's tools are exposed by this reusable tool definition. Leave empty to expose all of them; select specific tools to narrow the set. |
After you save and the connection succeeds, TrueServe discovers the server's tools so you can filter them and bind parameters.
When you attach an MCP tool to an agent, expand that selected tool in the agent's tool picker to enable only the operations that agent should use. The agent can only choose from operations already allowed by the MCP tool definition. Use the pop-out icon on the selected tool row when you need more room for operation choices and parameter bindings.
Parameter bindings¶
You can pre-fill a tool's parameters with fixed values that are hidden from the LLM — useful for tenant IDs, API scopes, or any constant the model shouldn't have to supply. Bound values can be literals, an $integration_id placeholder (resolved to the tool's integration), or a plain integration UUID. Anything you don't bind is left for the LLM to fill at call time.
For Salesforce Flow MCP tools, TrueServe unwraps the common inputs[] envelope and shows the flow's real fields directly. Bindings are still saved to the correct nested path behind the scenes.