CCOS MCP Tools

The CCOS MCP Tools suite provides a robust framework for interacting with the CCOS (Cognitive Cognitive Operating System) environment. It is designed for developers and autonomous agents that require governed, deterministic, and auditable tool execution. By leveraging the Model Context Protocol (MCP), these tools enable agents to discover available capabilities, manage complex, multi-step execution sessions, and perform memory-backed learning tasks. Whether you are building an AI agent that needs to search for external APIs, introspect remote endpoints, or execute plan-based logic in the RTFS (Runtime for Technical Functionality and Specification) language, these tools serve as the primary interface for secure agent autonomy. The system emphasizes clear boundaries between intent and execution, ensuring that every tool call, secret access, or plan modification is verified against a governance constitution.

• Capability Discovery: Utilize ccos_search, ccos_inspect_capability, and ccos_list_capabilities to dynamically find and validate tools registered within the CCOS ecosystem.
• Session Management: Seamlessly start, track, end, and consolidate sessions into reusable Agent Capabilities using session_start, session_end, and ccos_consolidate_session.
• Governed Execution: Execute JSON-based capabilities or raw RTFS code with ccos_execute_capability and ccos_execute_plan, ensuring all side effects pass through a strict governance kernel.
• Memory & Learning: Enhance agent performance by recording thoughts, retrieving memories with ccos_recall_memories, and documenting successful learning patterns.
• Development & Synthesis: Leverage advanced tooling like ccos_synthesize_capability for automated RTFS generation and ccos_compile_rtfs for syntax validation before runtime.
• Security & Governance: Manage system rules, check for available secrets, and monitor approval requests to ensure safe interaction with external servers and APIs.

Usage notes include: Always verify secret availability via ccos_check_secrets before executing sensitive operations. Use the session pattern (start -> execute -> consolidate) for complex tasks to maintain causal trails. The system is intended for environments where predictability and auditability are non-negotiable. Inputs typically involve JSON payloads for tool parameters, while outputs provide structured results or RTFS-compliant plan fragments. Constraints require that all agent behaviors are governed by the CCOS constitution and that RTFS code remains pure by default, with external effects isolated to explicit host calls.