AI导哈

CBT协议构建工具

Cerina Protocol Foundry是一个基于LangGraph和FastAPI的后端服务，结合React仪表板和MCP接口，用于构建需要人工审批的CBT协议。

  • by Silaskingfs
  • Python
  • 本地部署
  • 更新于
在 GitHub 查看 dev-saif-777/ProtocolFoundry

使用说明

Cerina Protocol Foundry

LangGraph + FastAPI backend with a React dashboard and MCP interface for building CBT protocols with human-in-the-loop approval.

Architecture (high level)

flowchart LR
  subgraph Interfaces
    REST["REST (FastAPI)"] -->|run/approve/history| API
    WS["WebSocket"] -->|node_update/state_update| API
    MCP["MCP (stdio tool)"] -->|build_cbt_protocol| API
  end

  subgraph API["FastAPI + LangGraph"]
    IG[intent_guard]
    IN[intake]
    DR[drafter]
    SA[safety]
    CR[critic]
    SV[supervisor]
    FI[finalize]
    HR[human_review]
    CP[(Checkpointer\nPostgres/SQLite)]
  end

  REST --> IG
  WS --> IG
  MCP --> IG

  IG -->|relevant| IN
  IG -->|out-of-scope| OUT["Out-of-scope response"]
  IN --> DR --> SA --> CR --> SV --> FI --> HR
  HR -->|approve| END1["Completed"]
  HR -->|reject| DR
  SV -->|revise| DR
  SV -->|finalize| FI

  CP --- IG
  CP --- IN
  CP --- DR
  CP --- SA
  CP --- CR
  CP --- SV
  CP --- FI
  CP --- HR

  subgraph DB["Persistence"]
    RUNS[(runs & run_events)]
  end

  API --> RUNS
  HR -->|pending_interrupt| RUNS

Backend (cbt_backend)

  • Agents: intent_guard → intake → drafter → safety → critic → supervisor → finalize → human_review (halts for approval; resume applies edits or loops back).
  • State: drafts, reviews, supervisor decision, metrics, scratchpad, final payload, human feedback, pending interrupts.
  • Persistence: checkpoints via langgraph-checkpoint-* (Postgres/SQLite); run metadata/events + pending_interrupt column ensured on startup.
  • Interfaces: REST (sessions, run, approve, runs/events), WS (live updates), MCP stdio tool build_cbt_protocol.

Backend setup

cd cbt_backend
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# env (example)
export DATABASE_URL=postgresql://postgres:postgres@localhost:5432/cbt?sslmode=disable
export CHECKPOINT_BACKEND=postgres
export OPENAI_API_KEY=...
export OPENAI_MODEL=gpt-4o-mini

# optional postgres
docker-compose up -d postgres

# run API
uvicorn app.main:app --reload --port 8000

# run MCP (stdio)
python3 -m mcp_server.server

Backend testing

cd cbt_backend
export DATABASE_URL=postgresql://postgres:postgres@localhost:5432/cbt?sslmode=disable
pytest tests/test_persistence.py

Frontend (cbt-frontend)

  • React/Vite dashboard with live WS pipeline view, pending approval actions, scratchpad, events log, and PDF export of finalized protocols.
  • REST is used as fallback (Refresh button) when WS is unavailable.

Frontend setup

cd cbt-frontend
npm install
npm run dev

Set env if needed: VITE_API_BASE_URL (default http://127.0.0.1:8000), VITE_WS_BASE_URL (default ws://127.0.0.1:8000).

MCP usage

  • From cbt_backend: python3 -m mcp_server.server
  • Connect via MCP Inspector or Claude Desktop custom MCP server (stdio command above) and call build_cbt_protocol with your prompt. auto_approve_on_halt can finalize halted runs automatically; set require_human_approval to enforce halts.

相关 MCP