Skip to main content

MCP

Bridge your agents to any Model Context Protocol server — call its tools, read its resources, and expand its prompts as ordinary Autessa tools.

All three types point at a server registered under MCP servers via mcpServerId. Autessa speaks JSON-RPC over HTTP (plain JSON or SSE responses), re-reads the server's auth configuration on every call so rotated tokens pick up immediately, and applies SSRF protection to the server URL. Requests time out after 30 seconds with automatic retries. The easiest way to create these tools is the create-from-MCP flow, which imports a server's advertised tools with their schemas in one step.

MCP Tool Call TOOLMCP​

Calls one named tool on the MCP server. The runtime arguments the agent supplies are type-coerced against mcpToolInputSchema — integers, numbers, booleans, arrays, and objects are converted from strings to the JSON types the schema declares — and sent as the tool-call arguments.

The result returned to the agent is the text content of the MCP tool's response; image or binary content blocks in the response are dropped. Failures come back as {"error": true, "errorCode": ..., "errorMessage": ...}.

Example​

Tool config
{
  "toolType": "MCP",
  "mcpServerId": "V1StGXR8Z5",
  "mcpToolName": "search_issues",
  "mcpToolInputSchema": "{\"type\":\"object\",\"properties\":{\"query\":{\"type\":\"string\"},\"limit\":{\"type\":\"integer\"}},\"required\":[\"query\"]}"
}

Configuration fields

  • Name
    mcpServerId
    Type
    string (ID)
    Description
    Required string (ID) value. References an MCP server by ID.
  • Name
    mcpToolName
    Type
    string
    Description
    Required string value.
  • Name
    mcpToolInputSchema
    Type
    string
    Description
    JSON schema for the tool's input parameters

MCP Resource TOOLMCP_RESOURCE​

Reads a single resource from the MCP server by URI. Text resources are returned to the agent as-is; binary resources come back as a base64 payload labeled with the mime type. Only the first content element of the resource is read.

Example​

Tool config
{
  "toolType": "MCP_RESOURCE",
  "mcpServerId": "V1StGXR8Z5",
  "resourceUri": "file:///reports/latest-summary.md",
  "resourceMimeType": "text/markdown"
}

Configuration fields

  • Name
    mcpServerId
    Type
    string (ID)
    Description
    Required string (ID) value. References an MCP server by ID.
  • Name
    resourceUri
    Type
    string
    Description
    Required string value.
  • Name
    resourceMimeType
    Type
    string
    Description
    Required string value.

MCP Prompt TOOLMCP_PROMPT​

Invokes a named prompt on the MCP server, passing the agent's runtime arguments as the prompt arguments. The server assembles its prompt messages and the tool returns them flattened as readable text, one **role**: content block per message, which the agent can then act on.

Example​

Tool config
{
  "toolType": "MCP_PROMPT",
  "mcpServerId": "V1StGXR8Z5",
  "promptName": "triage_ticket",
  "promptArgumentsSchema": "{\"type\":\"object\",\"properties\":{\"ticketId\":{\"type\":\"string\"}},\"required\":[\"ticketId\"]}"
}

Configuration fields

  • Name
    mcpServerId
    Type
    string (ID)
    Description
    Required string (ID) value. References an MCP server by ID.
  • Name
    promptName
    Type
    string
    Description
    Required string value.
  • Name
    promptArgumentsSchema
    Type
    string
    Description
    JSON schema for the prompt's arguments