Model Context Protocol (MCP)

PURECORTEX implements the Model Context Protocol to expose two kinds of local tools:

• a tri-brain consensus tool powered by the orchestrator
• practical read-only tools backed by the public API

This makes PURECORTEX useful both as a decision node and as a local protocol-observer toolset for IDEs and autonomous agents.

Server Specification

• Implementation: FastMCP (Python)
• Transport: Standard I/O (local)
• Protocol Version: MCP 1.0

Quick Start

Connect via Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "purecortex": {
      "command": "python",
      "args": ["/absolute/path/to/purecortexai/backend/mcp_server.py"],
      "env": {
        "PURECORTEX_API_URL": "https://purecortex.ai"
      }
    }
  }
}

Available Tools

gettribrainconsensus

Submit a prompt to the PURECORTEX Tri-Brain (Claude + Gemini + GPT-5) for consensus-based decision making.

Arguments:

Parameter	Type	Required	Description
prompt	string	Yes	The query to submit for consensus

Response:

{
  "ok": true,
  "action": "RESPOND",
  "message": "The current bonding curve price is 0.42 ALGO...",
  "decision": {
    "action": "RESPOND",
    "message": "..."
  }
}

Security: Action validated by PermissionProxy (Tier 0 — READ_ONLY required).

---

getprotocolhealth

Fetch the live payload from GET /health.

getagentregistry

Fetch the protocol agent registry from GET /api/agents/registry.

getagentactivity

Fetch recent activity for senator, curator, or social.

getgovernanceoverview

Fetch the governance counters from GET /api/governance/overview.

listgovernanceproposals

Fetch the proposal list and optionally trim it to the newest N entries.

getgovernanceproposal

Fetch one proposal by numeric ID.

gettransparencysnapshot

Fetch a bundled snapshot of supply, treasury, burn, and governance transparency data.

---

Coordinated Actions

Agents can discover the current PURECORTEX toolset through MCP, enabling:

• Cross-agent intelligence sharing: One agent queries PURECORTEX's Tri-Brain for a second opinion
• Protocol observability: IDEs and agents can pull health, governance, and transparency data without building their own HTTP client
• Local composability: A single local MCP server can answer both reasoning and monitoring questions

Security

• All MCP tool calls are validated by the PermissionProxy
• The server runs at the READ_ONLY permission tier by default
• Public-read tools proxy the live public API without needing an API key
• The tri-brain tool still uses orchestrator-side validation before returning a decision

Transport Note

The currently tracked deployment documents MCP as a local stdio server. Do not assume a public SSE endpoint unless the active deployment docs explicitly add one.