Skip to content

CRPv6 - MCP Compatibility

CRPv6 is designed to complement the Model Context Protocol (MCP). MCP standardizes how LLMs discover and call tools; CRP standardizes how agents are positioned on the right task with the right context. The two are complementary.

CRP as an MCP server

CRP can expose its capabilities as an MCP server so any MCP client (Claude Desktop, Cursor, Windsurf, etc.) can:

  • Query the Contextual Knowledge Fabric (CKF).
  • Run governed tool calls through the CRP capability router.
  • Retrieve facts, audit trails, and provenance evidence.

Planned server capabilities

Capability MCP tool name What it does
Knowledge query crp_query_knowledge Semantic + graph search over CKF
Tool execution crp_execute_capability Run a CRP capability with policy enforcement
Audit export crp_export_audit Return HMAC-signed audit events
Provenance check crp_check_provenance Verify claim-to-source attribution

Implementation target

  • Directory: crp/mcp/ (new)
  • Transport: stdio and SSE
  • Mapping: ToolCapabilityFabric → MCP tool definitions; CapabilityExecutor → MCP tool calls

CRP as an MCP client

CRP agents can consume any MCP server. This means a crp.Agent can use:

  • Filesystem servers
  • Database servers
  • Web search servers
  • GitHub servers
  • Any custom MCP tool

Planned adapter

  • File: crp/tools/adapters.py extended with MCPAdapter
  • Behavior: discover tools from MCP server, convert schemas to CRP ToolSpec, route calls through policy + audit

Why this matters

MCP solves tool access. CRP solves tool context - what the agent knows, what it should do next, and how to prove it was safe. Together they let a small local model use a universe of tools without drowning in schemas.

TODO

# Task Where
D1 Implement MCP server crp/mcp/
D2 Implement MCP client adapter crp/tools/adapters.py
D3 Example: CRP agent using filesystem + web MCP servers examples/mcp/
D4 Document authentication and sandboxing This page