# Frisky GPT MCP Beta

Frisky GPT MCP Beta is the official Frisky Dev private ChatGPT Developer Mode MCP connector at https://mcp.friskydev.com/mcp.

It is not an AI model provider by itself. It is a Model Context Protocol server that exposes Frisky specialist tools, a visual specialist console, and MongoDB-backed unified memory.

Secret/runtime policy:
- Infisical vault is the canonical secret source for Frisky dev/operator environments.
- Coolify is the dev runtime.
- Coolify should receive server-side runtime secrets from Infisical.
- Do not expose machine tokens, database credentials, MongoDB URI, Supabase service-role key, or Vertex service configuration in browser clients or public GPT instructions.

Primary public pages:
- Specialist gallery: https://mcp.friskydev.com/specialists
- Integration hub: https://mcp.friskydev.com/integrations
- Integration JSON manifest: https://mcp.friskydev.com/integrations.json
- CLI manifest: https://mcp.friskydev.com/cli.json
- CLI download: https://mcp.friskydev.com/frisky-mcp-cli.sh
- Specialist JSON manifest: https://mcp.friskydev.com/specialists.json
- Model/API notes: https://mcp.friskydev.com/model
- Machine-readable model/API manifest: https://mcp.friskydev.com/model.json
- GPT Actions OpenAPI schema: https://mcp.friskydev.com/openapi.json
- Codex MCP config: https://mcp.friskydev.com/codex.mcp.json
- MCP endpoint for ChatGPT Developer Mode: https://mcp.friskydev.com/mcp

Best supported API for default live specialist execution:
- OpenAI Responses API
- Environment variable: OPENAI_API_KEY
- Server model variable: FRISKY_GPT_MODEL
- Current recommended Frisky MCP model setting: gpt-5.2

Google Cloud mind lane:
- Vertex AI Gemini generateContent
- Select with FRISKY_MODEL_PROVIDER=vertex
- Project variable: VERTEX_AI_PROJECT_ID or GOOGLE_CLOUD_PROJECT
- Location variable: VERTEX_AI_LOCATION
- Model variable: VERTEX_AI_MODEL
- Current recommended Vertex model setting for Telegram credit efficiency: gemini-2.5-flash-lite
- Auth: Cloud Run service account metadata token with Vertex AI permissions

Telegram summon path:
- Telegram bot can call POST https://mcp.friskydev.com/api/consult
- GPT Actions can call POST https://mcp.friskydev.com/consult_specialist
- GPT Actions OpenAPI import URL: https://mcp.friskydev.com/openapi.json
- For live calls, send Authorization: Bearer ${FRISKY_BOT_API_TOKEN}
- Body: {"slug":"frisky-director","request":"Prioritize today's beta work from Telegram.","live_call":true}
- The backend picks OpenAI or Vertex according to FRISKY_MODEL_PROVIDER and configured credentials.

CLI path:
- Install: curl -fsS https://mcp.friskydev.com/frisky-mcp-cli.sh -o frisky-mcp && chmod +x frisky-mcp
- List specialists: ./frisky-mcp specialists
- Prompt packet: ./frisky-mcp packet frisky-codex-engineer "Review deployment risk"
- Live call: FRISKY_BOT_API_TOKEN=... ./frisky-mcp consult frisky-director "Prioritize beta work"

Beta status:
- Specialist discovery: supported
- Unified memory: supported through MongoDB
- Text fallback: supported through list_specialists
- Visual console: experimental across ChatGPT clients
- Live specialist execution: requires OPENAI_API_KEY or Vertex AI env configuration

Supported text fallback commands:
- Consult Frisky Director: prioritize today's beta work.
- Consult Frisky Launch Strategy: build a beta rollout checklist.
- Consult Frisky Codex Engineer: review deployment risk.