The Model Context Protocol (MCP) Registry manages tools that your agent can call. Orithos scans test these tools for security vulnerabilities.
Each MCP tool has a name, description, input schema (JSON Schema), and risk level. Tools are registered via the API and can be assigned to agents.
MCP tools communicate via JSON-RPC 2.0 protocol. Each tool call is a JSON-RPC request with method, params, and id fields. Responses include result or error with standard JSON-RPC error codes.
MCP tool calls are rate-limited to 120 requests per minute per agent. Rate limit headers are included in every response. Exceeding the limit returns a 429 status with a Retry-After header.
All tool inputs are validated against their registered JSON Schema before execution. Invalid inputs receive a detailed validation error identifying the specific field and constraint violation. This prevents injection attacks at the schema level.