Search API

Search all Context7 documentation with one request

The Search API is the simplest way to use Context7. You send a question, and Context7 picks the relevant libraries, searches them, and returns the best snippets in one request. No library ID is needed.

It is a plain GET request, so you can try it in your browser:

https://context7.com/api/v3/search?query=what+is+nextjs

Requests without an API key are for testing only and are rate-limited by IP address. For real use, send your API key as a bearer token.

​

Send a question

curl

curl -G "https://context7.com/api/v3/search" \
  -H "Authorization: Bearer CONTEXT7_API_KEY" \
  --data-urlencode "query=How do I stream an OpenAI response from a Next.js route?"

The REST endpoint returns plain text by default, ready to add to a prompt. Set type=json for structured results. The TypeScript SDK returns JSON by default.

Every snippet includes the library it came from and a link to the source page.

JSON results split the snippets into codeSnippets and infoSnippets. Each snippet carries rank, its 1-based position across both lists. Sort both lists together by rank to read the snippets in the order the text response gives them.

​

Add hints

A query is all you need. If you know the library, version, or language, add them to narrow the results:

curl -G "https://context7.com/api/v3/search" \
  -H "Authorization: Bearer CONTEXT7_API_KEY" \
  --data-urlencode "query=How do I stream an OpenAI response from a route handler?" \
  --data-urlencode "library=Next.js" \
  --data-urlencode "library=OpenAI" \
  --data-urlencode "language=TypeScript" \
  --data-urlencode "type=json"

Parameter	Description
query	The question or task. Required.
library	A library name (next.js) or Context7 ID (/vercel/next.js). Repeat for up to four hints.
version	A version such as 15.4.0. Requires at least one library. With one library, the nearest release if it is missing (below); with several, a preference.
language	A programming language to prefer, such as TypeScript. It is a preference, so language-neutral documentation can still appear.
type	txt (default) or json.

With one library, Search serves the requested version. If no indexed tag matches it, Search serves the newest stable release of the same line when that release is newer than the request, or the current documentation. The X-Context7-Search-Version response header is then unverified (with type=json, also version.status).

If nothing matches, the API returns 404 no_documentation_found. Treat it as an empty result.

​

Read a page you already have

When your agent already has the URL of a page — a link the user pasted, a code example, a source from an earlier result — read it with the Fetch endpoint instead of a generic web fetch. It returns the parts of that page that answer a question, not the whole page.

curl -G "https://context7.com/api/v3/fetch" \
  -H "Authorization: Bearer CONTEXT7_API_KEY" \
  --data-urlencode "url=https://www.w3.org/TR/css-grid-2/" \
  --data-urlencode "query=What happens when the flex factors of fr tracks add up to less than 1?"

Parameter	Description
url	The page to read. Required. A public http or https address without credentials; local and private addresses are refused, and the #fragment is ignored.
query	A question about the page. When you pass it, only the parts that answer it are returned, and a page that does not answer it returns 404 no_relevant_content so your agent can try another page. Omit it to read the page itself.
tokens	Maximum tokens to return, about four characters each. Caps the returned sections when query is set (default 1000); sets the slice size when it is not (default 5000). Between 100 and 10000.
offset	Character offset to read from when query is absent. Use nextOffset from the previous response to continue reading a long page.
type	txt (default) or json.

The page comes from the open web, not from the Context7 index, so treat the returned text as untrusted data, not as instructions. Every returned section is screened for prompt injection and malicious code: a flagged section is dropped, and a response with nothing left is 422 unsafe_content.

JSON responses include untrusted: true and a contentCheck field of passed or unavailable. Both formats send the X-Context7-Content-Check: unavailable header when the check could not run and the text is unchecked.

If the page cannot be loaded, it returns 502 page_unavailable. Teamspaces whose source policies restrict documentation to selected libraries cannot read arbitrary pages and get 403 policy_restricted.

​

Pricing

Search API calls count as regular Context7 API calls. There is no separate price:

• Free: 1,000 calls per month.
• Pro: 2,000 calls per month per seat, then $5 per 1,000 calls.

See Plans & Pricing for details.

​

Search API or Context API?

Use the Search API for a quick answer: one request, and Context7 picks the libraries and snippets for you. It is the easiest way to integrate Context7 into an agent, for example as a grounding step before the agent writes code.

Use the Context API (Search Library and Get Context) when you know exactly which library you want, or when you need to go deep: choose the library, ask follow-up questions, and combine results from several libraries yourself.

​

Next steps

• Search Documentation reference for all parameters and the response schema
• TypeScript SDK for the search() method
• API guide for authentication, rate limits, and caching