MCP Servers

MCP (Model Context Protocol) servers let you connect external tools, resources, and prompts to Prospector Studio agents over a standardized HTTP protocol.

What Are MCP Servers?

MCP is an open protocol that allows AI applications to interact with external services. Each MCP server can expose:

• Tools — Executable functions (e.g., search the web, read a file, run a query)
• Resources — Data access endpoints (e.g., configuration, database records)
• Prompts — Pre-defined prompt templates with arguments

When an MCP server is connected to Prospector Studio, its tools become available to agents during conversations.

Browsing and Configuring

From the Studio UI you can:

1. Browse the catalog of available MCP servers
2. Configure server connections (endpoint URL, credentials)
3. View the tools each server exposes
4. Attach MCP server tools to specific agents

Built-In MCP Servers

Prospector Studio ships with several pre-configured MCP server integrations:

How It Works

User message → Agent → Tool Registry
                           ↓
                   MCP Tool selected
                           ↓
              HTTP POST (JSON-RPC) → MCP Server
                           ↓
                   Tool result returned
                           ↓
                Agent incorporates result

1. A user sends a message to an agent
2. The agent decides which tool to call based on the conversation
3. Prospector Studio sends a JSON-RPC request to the MCP server's HTTP endpoint
4. The server executes the tool and returns results
5. The agent uses the results to formulate its response

Configuration

MCP servers can be configured at the system level via .mcp.json:

{
  "mcpServers": {
    "matrix-workflows": {
      "type": "http",
      "url": "https://studio.example.com/workflows/mcp"
    }
  }
}

Servers can also be registered through the Studio UI or GraphQL API for runtime configuration.

MCP vs Connectors

MCP servers and Connectors both extend agent capabilities, but serve different use cases: