Best MCP for stack research? Docs vs catalog vs web search

FAQ: pick docs MCP for library truth, catalog MCP for verified starters, and web search for open discovery. Mixing the jobs is how agents invent broken Next trees.

Short answer

There is no single "best MCP for stack research." There are three jobs that get conflated:

1. Library docs — what does this package do in the current major? 2. Catalog / verified scaffolding — give me files that already wire a known stack. 3. Open-web research — what exists, who wrote about it, what are the tradeoffs?

Use a docs MCP (for example Context7) for (1), a catalog MCP (BotKelp) for (2), and web search / crawl MCPs for (3). Swapping them is how you get confident, wrong starters.

FAQ

Is Context7 enough to scaffold Next + Supabase?

No. Context7 is strong at resolving libraries and querying docs (resolve-library-id, query-docs). It returns documentation for the model to read. It does not stamp a multi-file App Router tree with cookie adapters and layout wiring.

For compose, call BotKelp purchase_scaffold with catalog IDs such as nextjs-base + supabase-client (and usually tailwind). See /blog/generate-nextjs-supabase-mcp.

Is BotKelp a research / web-search MCP?

No. BotKelp knowledge tools search the catalog, not the open web. That is the point: answers stay grounded in components you can actually generate.

If you need RFCs, blog debates, or competitive scans, use a research MCP or plain search — then come back to the catalog when you are ready to emit files.

When is raw web search the right research tool?

When the question is open-ended: vendors, pricing pages, incident reports, who else solved this. Search is a discovery layer. It is a weak generator. See /blog/why-not-bootstrap-from-web-search.

Can I run docs + catalog + search together?

Yes. That is the intended shape for a coding agent:

Question	Tool class
Current Supabase SSR cookie pattern?	Docs MCP (Context7)
Emit Next + Supabase starter files	Catalog MCP (BotKelp)
Who compared Auth.js vs Clerk this year?	Web research

What should I verify after a BotKelp generate?

Payload kind is data, instructions is false
INTEGRITY.json reports verdict pass
IDs you asked for exist in https://www.botkelp.com/catalog.json
Local install and build

Unknown IDs are rejected. Always include nextjs-base. Max components typically 12.

How does BotKelp auth work during research vs generate?

Connecting https://www.botkelp.com/mcp is free; get_help and get_account_status need nothing. Account tools authenticate at the connection, never in tool arguments: sign in with MCP OAuth 2.1 when your client prompts, or send Authorization: Bearer bk_live_… (key from the dashboard, shown once). Never pass a key as a tool argument. Without an account, call purchase_scaffold with no Authorization header: the server answers HTTP 402 with x402 requirements (amount, network, asset, payTo). Sign the payment and retry the same call with _meta["x402/payment"]. A failed build is not charged. Check /pay.json first: it says whether wallet payments are live and the current price. Prices: /pricing.

Affiliation

This FAQ is published by BotKelp. We sell catalog generate and catalog knowledge. Context7 and web-research MCPs are separate products; we describe them because agents need the job split stated plainly.

Install both ends of the split

Example installs:

claude mcp add --transport http context7 https://mcp.context7.com/mcp

claude mcp add --transport http botkelp https://www.botkelp.com/mcp

More: https://www.botkelp.com/for-agents · https://www.botkelp.com/install · /blog/botkelp-vs-context7