# Mintlify Index MCP server (/search-index/mcp)

<!-- agent-signals: reading_time_min: 2 · est_tokens: 647 · updated: 2026-09-23 -->
Related: [Connect Mintlify Index to your coding agent](/search-index/connect.md), [Generate SDK reference pages from doc-tool output](/api-playground/sdk-reference-setup.md)

The Mintlify Index MCP server gives AI tools read-only access to technical documentation and web context. Connect to it at:

```text
https://index.mintlify.com
```

You do not need to authenticate to the public server. See [Connect](/search-index/connect) for Claude Code and Cursor setup instructions.

<Note>
  The Index MCP searches across covered product documentation and the technical web. To search only the content on a specific Mintlify-hosted site, use that site's [search MCP server](/ai/model-context-protocol).
</Note>

## `context` tool [#context-tool]

Use `context` to research an implementation task and return compact, source-cited context in one call. The tool is read-only and can access the open web.

<ParamField body="query" type="string">
  Implementation question to research.
</ParamField>

<ParamField body="product" type="string">
  Product or company name to use as an additional retrieval hint.
</ParamField>

<ParamField body="includeDomains" type="string[]">
  Domains to include in retrieval. When set, retrieval limits results to these domains.
</ParamField>

<ParamField body="excludeDomains" type="string[]">
  Domains to exclude from retrieval.
</ParamField>

<ParamField body="tokenBudget" type="integer" default="3000">
  Maximum number of tokens to return. The maximum value is `6000`. Use the default for focused questions and a larger budget for complex, multi-part tasks.
</ParamField>

### Output [#output]

The tool returns Markdown-formatted context assembled from ranked sources. Each section includes a title, source URL, and the relevant source content.

```text
### Configure caching

Source: https://nextjs.org/docs/app/getting-started/caching-and-revalidating

Relevant source content appears here.

--------------------------------
```

The MCP server returns source material for the connected agent to use. The agent decides how to apply that context to your task.

## Rate limits [#rate-limits]

Index applies the following per-IP limits to the public MCP server:

| Window     |          Limit |
| ---------- | -------------: |
| Per second |    10 requests |
| Per day    | 5,000 requests |

Requests exceeding either limit return `429 Too Many Requests`. Wait before retrying, and use exponential backoff for automated clients.

## Protocol behavior [#protocol-behavior]

The Index MCP uses stateless Streamable HTTP:

* Send each JSON Remote Procedure Call (JSON-RPC) request as an HTTP `POST`.
* Send one JSON-RPC request per HTTP request. Batching is not supported.
* Do not persist or send an MCP session ID between requests.