quilt-canon-mcp

MCP (Model Context Protocol) server exposing Quilt substrate walker canon as tools for AI agents.

Why this matters

Other AI agents (Claude, GPT, etc.) that support MCP can now directly query our canon — every doctrine, every cell, every JEV verification. The canon becomes a tool in someone else's toolkit.

Related MCP server: okfy

Quick start

Run the MCP server

pip install -e .
quilt-canon-mcp
# Server listens on stdio. Send JSON-RPC 2.0 requests.

Test via CLI

quilt-canon-mcp-cli initialize
quilt-canon-mcp-cli tools/list
quilt-canon-mcp-cli tools/call --tool list_doctrines
quilt-canon-mcp-cli tools/call --tool search_canon --args '{"query": "scar witness", "top_k": 3}'
quilt-canon-mcp-cli tools/call --tool probe_canon --args '{"lore_text": "The cell is a scar..."}'

Use from any MCP client

import json, subprocess

proc = subprocess.Popen(
    ["quilt-canon-mcp"],
    stdin=subprocess.PIPE, stdout=subprocess.PIPE, text=True
)

req = {"jsonrpc": "2.0", "id": 1, "method": "tools/call",
       "params": {"name": "list_doctrines"}}
proc.stdin.write(json.dumps(req) + "\n")
proc.stdin.flush()
print(proc.stdout.readline())

Tools exposed

Resources exposed

• canon://list — all canon pieces

• canon://doctrines — the 5 bedrock doctrines

Fleet integration

This tool uses the existing fleet as acceleration:

• quilt-multi-oracle — probe_canon() falls back to local heuristic; if DEEPINFRA_TOKEN is set, calls DeepInfra Llama-3.3-70B.

• quilt-canon-graph (sibling) — knowledge graph of canon

• quilt-canon-search (sibling) — TF-IDF search

• quilt-iterator — reference for canon structure

• quilt-bridge — translates canon across substrates

The 5 bedrock doctrines

1. cells_are_scars — every cell records an attempted entry

2. witness_log_is_prediction — the log IS the prediction

3. canon_gate_is_chord — canon passes when multiple agents agree

4. oracle_is_heard — JEV probes canon with multi-model consensus

5. substrate_quantum — the substrate is the walker; canon is substrate-aware

Polyformalism canary

python -m quilt_canon_mcp.canary
# → 0x024a555471370b18d

License

MIT