BotKelp vs Context7: scaffolds vs live library docs

Context7 pulls current library docs into the agent. BotKelp stamps verified Next.js catalog source. Complementary tools — use both, do not swap them.

Context7 and BotKelp both show up in MCP install lists for coding agents. They answer different questions.

Context7 is a docs MCP: resolve a library, then query current documentation so the model is not guessing from training cutoffs.

BotKelp is a scaffold catalog MCP: pick stable component IDs, call purchase_scaffold, and get stamped source the server already stores — not a prose description of how to wire Supabase.

If you treat either as a substitute for the other, you get the wrong failure mode.

What each tool actually does

	Context7	BotKelp
Job	Live library documentation	Compose verified Next.js starters
Typical tools	`resolve-library-id`, `query-docs`	`get_help`, `get_component`, `purchase_scaffold`
Remote HTTP	`https://mcp.context7.com/mcp`	`https://www.botkelp.com/mcp`
Output shape	Doc excerpts for the model	Files + `INTEGRITY.json`
Auth shape (common)	Optional API key for higher limits (often via header)	Connection free; account tools use MCP OAuth 2.1 or `Authorization: Bearer bk_live_…`, never a tool argument

Soft numbers only: Context7 commonly offers a free tier with an API key for higher limits. Confirm pricing and limits on Context7's own site — do not hardcode third-party prices from blog posts.

Claude install examples

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

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

Keep both. Agents that research *and* scaffold need both slots filled.

When Context7 is the right call

"What changed in Next.js App Router cookies helpers this major?"
"Show the current Prisma client constructor pattern for this version."
"Resolve the canonical library id, then quote the docs."

You want authoritative prose and examples from docs, not a zip of starter files.

When BotKelp is the right call

"Give me a Next + Tailwind + Supabase tree that already builds."
"Compose nextjs-base, tailwind, supabase-client — do not invent the cookie adapter."
"Return kind: data, instructions: false, and an INTEGRITY pass."

You want named catalog blocks stamped into destinations, not a model improvising createBrowserClient from memory.

Always include nextjs-base. Unknown IDs are rejected. Max components is typically 12. IDs live at catalog.json.

Complementary, not competing

A sane agent loop:

1. Context7 (or official docs) for *how the library works today* 2. BotKelp for *the starter files that already wire those patterns* 3. Your agent for the product code that does not exist yet

BotKelp knowledge tools search the catalog, not the open web. Asking them for "latest React RFC" is the wrong tool. Asking Context7 to emit a stamped multi-file Supabase App Router tree is also the wrong tool.

Affiliation

This post is published on BotKelp's site. We sell scaffold generation and catalog knowledge. Context7 is a separate product from Upstash / the Context7 project. We are not affiliated with Context7; we recommend it because docs MCP and catalog MCP solve different jobs.

Try BotKelp

Connecting the MCP URL is free and unauthenticated:

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

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.

More: /install · /for-agents · /llms.txt