MCP server for coding agents
The honest second opinion your agent asks before it builds.
Tell your agent what you're making. It asks Cuely which tools to use and how to set them up, and gets a stack with reasons, real free-plan limits and step-by-step recipes. Free, no account, and no vendor can pay to be the answer.
https://cuely.dev/api/mcpyouI'm building a SaaS for a fintech startup. Use Cuely for the stack.
agentcuely.recommend_stack("SaaS for a fintech startup")
- HostingVercelFirst-class Next.js support with a preview URL for every branch.
- DatabaseNeonSOC 2 and HIPAA (with a BAA) are listed on the Scale plan.
- AuthWorkOS AuthKitOrganizations are the core model, and enterprise SSO is one switch away.
- PaymentsStripeInvoices, seats, usage billing and a hosted customer portal.
- EmailPostmarkActivity retention up to 365 days on Pro helps audits and support.
- StorageAmazon S3Mature controls: object lock, encryption keys, access logs.
cuelyMatched recipe b2b-saas. Next: get_recipe for the build steps.
How it works
- 01
Your agent describes the build
One or two plain sentences: who it's for, what it does, and any constraints like budget, region, compliance or tools you already use.
- 02
Cuely matches needs to published rules
Words like B2B, fintech, EU or self-host become needs. Each tool has a published score for each need. The same question gets the same answer, for everyone.
- 03
The agent gets a plan it can follow
A pick per category with the reason, the free plan's actual limits, what to watch out for, and a full recipe when one fits.
Connect in one step
Cuely speaks MCP over Streamable HTTP. Add it once and any session can use it. Then just ask: Use Cuely to choose the stack for this.
Run once in your terminal.
claude mcp add --transport http cuely https://cuely.dev/api/mcpFour tools, all read-only
recommend_stack(description, categories?, already_using?, avoid?)One pick per category with a one-line reason, the free plan's real limits, and the alternatives worth knowing.
get_recipe(product_type, include_steps?)The full build: ordered steps, commands, starter files, best practices, pitfalls and compliance notes.
compare_tools(a, b, context?)Two options in one category, side by side, with a verdict for what you're building.
list_categories(category?)Browse the catalog: every tool's free plan, starting price and the page those facts came from.
The catalog
30 tools across 10 categories, plus the b2b saas with auth, billing and email recipe. Every free plan and price was read from the vendor's own site on Nov 6, 2026.
Hosting
Where the app's frontend and server code run.
- Vercel
- Cloudflare Workers
- Netlify
- Render
Database
The primary datastore for application data.
- Neon
- Supabase
- Turso
Auth
Sign-up, sign-in, sessions, and organizations.
- Clerk
- Better Auth
- WorkOS AuthKit
- Supabase Auth
- Neon Auth
Payments
Checkout, subscriptions, invoices, and tax.
- Stripe
- Paddle
- Lemon Squeezy
Email
Transactional email: invites, receipts, password resets.
- Resend
- Postmark
- Amazon SES
Secrets
API keys and environment variables across environments.
- Infisical
- Doppler
DNS
Domain records for the app and for email authentication.
- Cloudflare DNS
- Amazon Route 53
Monitoring
Error tracking, uptime checks, and logs.
- Sentry
- Better Stack
Analytics
Product and web analytics.
- PostHog
- Vercel Web Analytics
- Plausible
Storage
File and object storage for uploads and exports.
- Cloudflare R2
- Vercel Blob
- Amazon S3
Money never touches the ranking.
Cuely's picks come from published rules in its open catalog. No vendor can pay for placement, and there are no affiliate links.
- Published rules. Each tool has a score from −3 to +3 for each need, with the reason written next to it. A pick is the sum of the needs you described, plus a small editorial default that only settles close calls.
- Same answer for everyone. No accounts, no personalization, no live model call deciding your stack. The same description returns the same stack.
- Facts from the source. Free plans and prices come from each vendor's own pricing page, with the date and the page in every answer. When a page is silent, Cuely says so instead of guessing.
- What you already use counts. If a tool you have still fits, Cuely keeps it and tells you switching isn't worth it.
score = Σ fit[need] + 0.4 × (n − default_rank + 1) + (already_using ? 4 : 0)