Remote MCP surface
Use this for ChatGPT Developer Mode, MCP-aware agents, and tool clients that support streamable HTTP.
https://mcp.friskydev.com/mcp
Frisky Dev Cloud
MCP Integration Hub
One accelerated builder ecosystem for turning ideas into running Frisky tools: MCP clients, GPT Actions, REST automations, Codex configs, Telegram, and agent runners.
MCP Clients
Codex, Cursor, Antigravity, and ChatGPT all point at the same remote MCP endpoint.
MCP server JSON
Install the repo plugin at plugins/frisky-dev-mcp, or copy this config into MCP-capable clients that accept remote server URLs.
{
"mcpServers": {
"frisky-dev-mcp": {
"type": "http",
"url": "https://mcp.friskydev.com/mcp"
}
}
}
Cursor MCP config
Use the same remote MCP bridge for coding sessions, reviews, and specialist planning inside Cursor.
https://mcp.friskydev.com/cursor.mcp.json
Antigravity MCP config
Use Frisky MCP as the Antigravity cloud tool surface for product, infrastructure, launch, and execution routing.
https://mcp.friskydev.com/antigravity.mcp.json
Private repo cards
Behind the Frisky login gate, specialists can answer from allowlisted private GitHub repos using a server-side read token. Secret-like files are skipped and secret-looking values are redacted.
FRISKY_GITHUB_TOKEN=github-fine-grained-token FRISKY_GITHUB_REPOS=FriskyDevelopments/frisky-dev-mcp-plugin
Temporary GitHub token path
If vault access is blocked, set the GitHub token directly in the gated server runtime. For local testing, use the already authenticated GitHub CLI token without writing it into repo files.
export FRISKY_GITHUB_TOKEN="$(gh auth token)" export FRISKY_GITHUB_REPOS="FriskyDevelopments/frisky-dev-mcp-plugin" npm run smoke
Actions + REST
GPTs, bots, and services can call the same specialist action without embedding secrets in prompts.
OpenAPI import
Import this schema in GPT Builder. Configure API Key authentication as Bearer using your Frisky bot token.
https://mcp.friskydev.com/openapi.json
Consult specialist
Use a Frisky OAuth access token from the MCP login flow for human/interactive calls, or a machine bearer token for server jobs.
POST https://mcp.friskydev.com/consult_specialist
Authorization: Bearer $FRISKY_ACCESS_TOKEN_OR_BOT_TOKEN
Content-Type: application/json
{
"slug": "frisky-director",
"request": "Prioritize today's beta work.",
"live_call": true
}
CLI
A tiny shell runner for local machines, CI, Coolify dev containers, and ops scripts.
Terminal support
Download the CLI, inspect it, then run text-first packet or live consult commands.
curl -fsS https://mcp.friskydev.com/frisky-mcp-cli.sh -o frisky-mcp chmod +x frisky-mcp ./frisky-mcp specialists
Packet and live call
Packet mode needs no machine token. Live consults use a Frisky OAuth access token or server bot token.
./frisky-mcp packet frisky-codex-engineer "Review deployment risk" FRISKY_ACCESS_TOKEN=... ./frisky-mcp consult frisky-director "Prioritize beta work" FRISKY_BOT_API_TOKEN=... ./frisky-mcp consult frisky-director "Prioritize beta work"
Backend Integrations
Database and edge runtimes call specialists from server-side code only.
Database agents
Neon-backed agents can call Frisky Data Model, SaaS Infrastructure, or Codex Engineer through the REST action endpoint.
POST https://mcp.friskydev.com/consult_specialist
Authorization: Bearer $FRISKY_BOT_API_TOKEN
{
"slug": "frisky-data-model",
"request": "Review this schema migration.",
"live_call": true
}
Edge Function call
Supabase Edge Functions can call the same live specialist endpoint with bearer auth stored as a project secret.
await fetch("https://mcp.friskydev.com/consult_specialist", {
method: "POST",
headers: {
Authorization: `Bearer ${Deno.env.get("FRISKY_BOT_API_TOKEN")}`,
"Content-Type": "application/json"
},
body: JSON.stringify({
slug: "frisky-trust-safety",
request: "Review this auth policy.",
live_call: true
})
})
Secrets + Dev Runtime
Use the vault you can reach: 1Password for operators, Cloudflare secrets for Workers, and Google Cloud Secret Manager for Cloud Run. Browsers and GPT instructions get placeholders only.
Operator vault
Use a 1Password template when Infisical is blocked. The resolved file stays local and ignored by git.
op inject \ -i ops/secrets/frisky-mcp.env.tpl \ -o .env.local \ --force
Worker secrets
Use Cloudflare secrets for Workers that proxy, gate, or call Frisky MCP. Secret values stay hidden after upload.
npx wrangler secret put FRISKY_GITHUB_TOKEN npx wrangler secret put FRISKY_BOT_API_TOKEN npx wrangler deploy --secrets-file .env.production
Cloud Run secrets
Use Secret Manager for the live Cloud Run MCP service, then expose selected secrets as runtime env vars.
gcloud run services update frisky-gpt-mcp \ --region us-central1 \ --set-secrets="FRISKY_GITHUB_TOKEN=frisky-github-token:latest"
Recover vault path
When Infisical access returns, move temporary runtime values back into the canonical vault and rotate any broad temporary token.
1Password / GitHub CLI -> Cloudflare or Google runtime -> Infisical when access returns
Security rule: never place service-role keys, database credentials,
model API keys, or machine tokens in browser clients, public MCP configs, GPT instructions,
or screenshots. Use bearer placeholders in docs and inject real secrets server-side only.
If Infisical is unavailable, use a temporary host environment variable or GitHub CLI
token for the gated runtime, then move the value back into Infisical when access returns.