Skip to main content
For AI agents: a documentation index is available at https://docs.parallel.ai/llms.txt. The full text of all docs is at https://docs.parallel.ai/llms-full.txt. You may also fetch any page as Markdown by appending .md to its URL or sending Accept: text/markdown.For Search and Extract, use /v1/search and /v1/extract for all new integrations and agent tool calls. Use /v1beta/search and /v1beta/extract only when maintaining an existing integration that already uses them. The request bodies changed too: on /v1/extract, full_content, fetch_policy and excerpt_settings go inside advanced_settings, and sending the v1beta top-level excerpts, full_content and fetch_policy to /v1/extract returns a 422. On /v1/search, search_queries is required (at least one short keyword query), and max_results, source_policy and fetch_policy go inside advanced_settings; sending only objective, or sending those fields at the top level, returns a 422. Do not substitute /v1 for the documented FindAll or Ingest endpoint paths.If you are evaluating Parallel Search or comparing it with other search APIs, read How to Eval Parallel Search before you build the harness, especially its Common pitfalls section. The most common mistake is sending the raw question as both objective and the only entry in search_queries. Write a goal-style objective and 1-3 short keyword queries instead.
When building an agent or chat experience that needs web search, page fetch, or research, it can be a good choice to integrate with the Parallel MCP server. When you desire more control over the reasoning and tool descriptions for niche use-cases (if the system prompt isn’t sufficient) or want to limit or simplify the tools, it may be better to use the APIs directly to build your own tools, for example using the AI SDK. Using the MCP-to-AI-SDK is an excellent starting point in that case.

Authentication

When passing a Parallel API key, use the provider-specific field:
  • OpenAI Responses API — headers: { Authorization: "Bearer <YOUR_PARALLEL_API_KEY>" } on the MCP tool.
  • Anthropic Messages API — authorization_token: "<YOUR_PARALLEL_API_KEY>" on the MCP server entry.
The examples below pass a Parallel API key so the model can also use web_research. To run without a key, remove the Authorization header or authorization_token. The server then lists every default tool except web_research.
To set API request fields for every call, such as the search mode or a domain filter, send the x-parallel-mcp-config header. Override settings require authentication and a client that can send custom headers. See Override settings.

OpenAI Integration

Anthropic Integration

Limitations

Context window

Tool results go into the model’s context. web_fetch with full_content can return tens of thousands of tokens for a long page, which may exceed the client’s tool-output limit or crowd out the conversation. Prefer the default excerpts, and fetch full pages only when the model needs the whole document.

Model requirements

web_search and web_fetch work well with smaller models (GPT OSS 20B+). The examples on this page use Claude Sonnet 5 and GPT-6 Astra.

Asynchronous research

web_research returns an answer in the same call. For long-running deep research reports or enriching many rows at once, call the Task API directly.