Skip to content
CI

Guided connector setup, setup status, and setup from the AI client #2410

Sign in to view logs

Guided connector setup, setup status, and setup from the AI client

Guided connector setup, setup status, and setup from the AI client #2410

Workflow file for this run

.github/workflows/ci.yml at 6c2d7d3
name: CI
on:
push:
branches: [main]
pull_request:
branches: [main]
workflow_dispatch:
jobs:
backend:
name: Backend (lint, typecheck, test, build)
runs-on: ubuntu-latest
services:
postgres:
image: postgres:17-alpine
env:
POSTGRES_USER: amcp
POSTGRES_PASSWORD: testpassword
POSTGRES_DB: anythingmcp_test
ports:
- 5432:5432
options: >-
--health-cmd pg_isready
--health-interval 10s
--health-timeout 5s
--health-retries 5
env:
DATABASE_URL: postgresql://amcp:testpassword@localhost:5432/anythingmcp_test
# CI-only secrets — long enough to pass our boot validation, never used
# outside the test database.
JWT_SECRET: ci-test-jwt-secret-at-least-32-chars-aaaaaa
ENCRYPTION_KEY: ci-test-encryption-key-32-chars-aaaaaaaa
steps:
- uses: actions/checkout@v6
- uses: actions/setup-node@v6
with:
node-version: 22
cache: npm
- name: Install dependencies
run: npm ci
- name: Validate adapter catalog
run: node scripts/validate-adapters.mjs
- name: Test adapter validator diagnostics
run: node --test scripts/validate-adapters.test.mjs scripts/adapter-new.test.mjs
- name: Test the satellite generator
run: node --test scripts/satellites/satellites.test.mjs
- name: Verify the quoted adapter count and license wording
run: node scripts/adapter-count.mjs --check
# The Docker image ships only hoisted dependencies; a nested one passes
# every test here and then crashes the container (see the script).
- name: Verify no backend dependency is nested out of the image
run: |
node --test scripts/check-runtime-deps.test.mjs
node scripts/check-runtime-deps.mjs
- name: Verify catalog.ts is up-to-date with adapter JSONs
run: |
node scripts/regenerate-catalog.mjs
if ! git diff --quiet packages/backend/src/adapters/catalog.ts; then
echo "::error::catalog.ts is out of sync with adapter JSON files."
echo "Run: node scripts/regenerate-catalog.mjs && commit the result."
git diff packages/backend/src/adapters/catalog.ts
exit 1
fi
- name: Generate Prisma client
run: npx prisma generate
working-directory: packages/backend
- name: Run Prisma migrations
run: npx prisma migrate deploy
working-directory: packages/backend
- name: Lint
run: npm run lint
working-directory: packages/backend
- name: Type-check
run: npx tsc --noEmit -p tsconfig.json
working-directory: packages/backend
- name: Run tests
run: npm test
working-directory: packages/backend
- name: Build backend
run: npm run build
working-directory: packages/backend
frontend:
name: Frontend (lint, typecheck, build)
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- uses: actions/setup-node@v6
with:
node-version: 22
cache: npm
- name: Install dependencies
run: npm ci
- name: Lint
run: npm run lint
working-directory: packages/frontend
- name: Type-check
run: npx tsc --noEmit -p tsconfig.json
working-directory: packages/frontend
- name: Build frontend
run: npm run build
working-directory: packages/frontend
env:
NEXT_PUBLIC_API_URL: http://localhost:4000
release-script:
name: Cloud release script (blue/green)
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
# deploy/cloud/release.sh swaps the cloud's app containers blue/green;
# it only ever runs in anger in production. Exercise it on every PR,
# under load through Caddy: the migration from the single-container
# layout, healthy releases with zero failed requests, broken image,
# broken compose, wrong Caddyfile, a failure after the switch, repeated
# releases, the cold fallback — and nothing orphaned after any of them.
- name: Exercise release.sh against a throwaway stack
run: bash deploy/cloud/release.test.sh
stuck-users-report:
name: Cloud stuck-users report
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
# deploy/cloud/stuck-users-report.sh runs once a week on the droplet with
# SQL written against the production schema. Run it against every
# migration and fake data on each PR, so a schema change that breaks it
# fails here instead of as a Monday without a report.
- name: Exercise the weekly report against a throwaway database and SMTP
run: bash deploy/cloud/stuck-users-report.test.sh
docker:
name: Docker build
runs-on: ubuntu-latest
needs: [backend, frontend]
# Every push to main, and the pull requests that can break the image
# without breaking a unit test (decided in the first step). js-yaml 5
# (#809) passed every PR check and only failed here, after the merge.
if: (github.event_name == 'push' && github.ref == 'refs/heads/main') || github.event_name == 'pull_request'
steps:
- uses: actions/checkout@v6
with:
fetch-depth: 0
- name: Decide whether this change can break the image
id: scope
env:
EVENT: ${{ github.event_name }}
BASE_REF: ${{ github.base_ref }}
run: |
if [ "$EVENT" != "pull_request" ]; then
echo "run=true" >> "$GITHUB_OUTPUT"
exit 0
fi
changed=$(git diff --name-only "origin/$BASE_REF...HEAD")
if echo "$changed" | grep -qE '(^|/)package(-lock)?\.json$|^Dockerfile$|^\.dockerignore$|^start\.sh$|^packages/backend/prisma/|^packages/backend/prisma\.config\.ts$'; then
echo "run=true" >> "$GITHUB_OUTPUT"
else
echo "No dependency, Dockerfile or startup change: skipping the image boot."
echo "run=false" >> "$GITHUB_OUTPUT"
fi
- name: Build unified image
if: steps.scope.outputs.run == 'true'
run: docker build -t anythingmcp:ci .
# Building proves the image assembles, not that it runs. v0.8.0 shipped a
# backend that threw MODULE_NOT_FOUND at import time because a relative
# require resolved differently under dist/ than under src/: the image
# built clean, every test passed, and the container crash-looped. Boot it.
- name: Boot the image and wait for /health
if: steps.scope.outputs.run == 'true'
run: |
set -euo pipefail
docker network create amcp-ci
docker run -d --name amcp-ci-db --network amcp-ci \
-e POSTGRES_USER=amcp -e POSTGRES_PASSWORD=amcp -e POSTGRES_DB=anythingmcp \
postgres:17-alpine
# The entrypoint runs migrations and then starts the backend whether or
# not they succeeded, so starting both containers at once just means the
# app comes up against an empty schema. docker-compose gates on
# service_healthy; a bare docker run has to do it by hand.
for i in $(seq 1 30); do
docker exec amcp-ci-db pg_isready -U amcp -d anythingmcp -q && break
sleep 2
done
docker exec amcp-ci-db pg_isready -U amcp -d anythingmcp
# Booted the way the cloud runs it: OAuth2 on, and Sentry on. On
# 2026-09-24 Sentry's Express instrumentation renamed every middleware
# layer, @rekog/mcp-nest-auth's bootstrap check stopped finding
# cookie-parser, and the backend refused to start in production. The
# default boot (auth mode none, no DSN) loads neither module and
# could not have caught it. The DSN points at a .invalid host, so
# nothing is ever sent.
docker run -d --name amcp-ci-app --network amcp-ci -p 4000:4000 \
-e MCP_AUTH_MODE=oauth2 \
-e SENTRY_DSN=https://public@o0.ingest.sentry.invalid/0 \
-e DATABASE_URL=postgresql://amcp:amcp@amcp-ci-db:5432/anythingmcp \
-e JWT_SECRET="$(openssl rand -hex 32)" \
-e ENCRYPTION_KEY="$(openssl rand -hex 32)" \
-e NODE_ENV=production -e PORT=4000 \
-e CORS_ORIGIN=http://localhost:3000 \
-e SERVER_URL=http://localhost:4000 \
-e FRONTEND_URL=http://localhost:3000 \
anythingmcp:ci
for i in $(seq 1 60); do
if curl -sf http://localhost:4000/health > /dev/null; then
echo "backend healthy after ${i}0s"
exit 0
fi
sleep 10
done
echo "::error::the image built but the backend never became healthy"
docker logs amcp-ci-app 2>&1 | head -120
echo '--- last 60 lines ---'
docker logs amcp-ci-app 2>&1 | tail -60
exit 1
- name: Tear down
if: always()
run: |
docker rm -f amcp-ci-app amcp-ci-db 2>/dev/null || true
docker network rm amcp-ci 2>/dev/null || true