MCP servers that can search documentation

6 verified servers expose a tool that can search documentation and knowledge bases

When an agent hits an API it does not fully know, the fastest fix is to let it read the docs. A growing number of product MCP servers ship a documentation-search tool, so the model can look up the current, correct way to do something instead of guessing from stale training data.

These verified servers let an agent search official documentation or a knowledge base, scoped to that product.

Top pick

Cloudflare

Cloudflare

Official

Cloudflare's remote MCP servers: build and manage Workers, KV, R2, D1, and Hyperdrive from an agent.

cloud-infra3,806
Tool:
  • search_cloudflare_documentation

Workers, DNS, and caching config drift constantly, so search_cloudflare_documentation pulls the current reference instead of letting an agent configure against last year's API.

Pick 2

Stripe

Stripe

Official

Stripe's official MCP server lets agents create customers, payment links, invoices, and read balances.

payments1,583
Tool:
  • search_stripe_documentation

Looks up Stripe's documentation inline, which matters for a payments API where using a deprecated parameter is the difference between a charge and an error.

Pick 3

Supabase

Supabase (community)

Community

Connects agents to a Supabase project: run SQL, inspect schema, read logs, and manage edge functions.

databases2,710
Tool:
  • search_docs

RLS and client setup are easy to get subtly wrong, so search_docs sits next to the database tools and lets an agent check the right pattern while it builds against the project.

Pick 4

Sentry

Sentry

Official

Sentry's official MCP server: pull issues, stack traces, and events, and run Seer root-cause analysis from your editor.

monitoring-observability712
Tool:
  • search_docs

Setting up an SDK or an alert rule needs the exact config shape, which search_docs surfaces straight from Sentry's own reference.

Pick 5

MongoDB

MongoDB

Official

MongoDB's official MCP server: query and manage databases, plus Atlas cluster administration.

databases1,039
Tool:
  • search-knowledge

search-knowledge hits the MongoDB Assistant's curated knowledge base, so an agent writing an aggregation pipeline can confirm operator syntax instead of half-remembering it.

Pick 6

Apify

Apify

Official

Official Apify server that exposes 6,000+ Actors plus run, dataset, and store tools so agents can scrape and automate the web.

search-and-data1,300
Tool:
  • search-apify-docs

Assembling the right actor and input shape is the hard part of a scraping run, and search-apify-docs is where an agent looks that up.

What to know

This capability quietly kills one of the most common agent failures: confidently calling an API that changed or never existed. A doc-search tool grounds the model in the vendor's own current reference before it writes the call. Cloudflare and Stripe put that search in the same server as the tools that deploy a Worker or create a charge, so the agent can confirm the current parameter shape in the session it writes the call. Scope is the other half: search_docs on Supabase returns only Supabase, so an RLS question does not surface a four-year-old answer for a different version.

It pays off most when the agent keeps what it reads. Re-deriving the same auth flow or rate-limit rule every session is slow and dull. Docs are the source of truth; an agent's working knowledge of a product is something to accumulate, not re-fetch cold on every run.

Questions

How is this different from searching the web?
Scope and trust. A documentation-search tool queries one vendor's own current reference, so the agent gets authoritative, version-correct answers without sifting blog posts and forum threads. Web search is broader but noisier. For using a specific product's API correctly, the scoped doc tool wins.
Does this stop an agent from hallucinating an API?
It helps a lot. Grounding the model in the vendor's real documentation before it writes a call replaces guessing with reading. It is not a guarantee, but an agent that checks the docs makes far fewer up-to-date-API mistakes than one working from memory alone.