> ## Documentation Index
> Fetch the complete documentation index at: https://docs.runpulse.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Claude Connector

> Add Pulse to Claude as a remote MCP connector with OAuth, no API key required.

> Add Pulse from the Claude connector directory and let Claude extract, structure, and
> split your documents in the conversation.

The **Pulse connector** is a hosted remote MCP server that lets Claude (and other MCP
clients like Claude Desktop) call Pulse's document-extraction API directly inside a
conversation. Add it once and Claude can parse documents, apply schemas, split by topic,
and pull tables as tools, deciding the steps itself.

The connector endpoint is:

```
https://mcp.runpulse.com/mcp
```

Authentication uses **OAuth 2.0**, so there is no API key to copy or paste. You authorize
Claude against your Pulse account once, and every call is authenticated as you and billed
to your Pulse organization's credits.

<Note>
  This page covers the **directory connector** (OAuth). To wire Pulse into Claude Code,
  Codex, or VS Code manually with a Pulse API key instead, see
  [Connecting a client](/mcp/connecting).
</Note>

## Prerequisites

* An active [Pulse account](https://platform.runpulse.com) with available credits. Usage
  through the connector draws on your Pulse organization's credit balance.

## Connect Pulse to Claude

<Steps>
  <Step title="Add the connector">
    In Claude, open **Settings → Connectors**, find **Pulse** in the directory, and click
    **Connect**.
  </Step>

  <Step title="Log in to Pulse">
    Claude redirects you to sign in to your Pulse account.
  </Step>

  <Step title="Approve access">
    Review the consent screen ("Claude wants to access Pulse…") and approve it. Claude is
    now connected, and the Pulse tools appear in your tool list.
  </Step>
</Steps>

That's the whole setup. Because the connector uses OAuth, there is no API key to manage:
your Pulse credits and limits apply automatically to every call Claude makes.

### Claude Desktop custom connector

If you add Pulse as a **custom connector** in Claude Desktop rather than from the
directory, the connector dialog also asks for an OAuth **Client ID** and **Client Secret**.
Use the Client ID `claude-connector` together with the Client Secret that Pulse provides.
Adding Pulse from the directory listing handles these for you automatically, so most users
never need to enter them.

## Tools

The connector exposes eight tools. Most work starts with `extract` and chains the returned
`extraction_id` into the others.

| Tool              | Display name            | What it does                                                             |
| ----------------- | ----------------------- | ------------------------------------------------------------------------ |
| `extract`         | Extract Document        | Extract content and markdown from a document via a file URL.             |
| `apply_schema`    | Apply Schema            | Apply a JSON schema against a prior extraction to get structured fields. |
| `generate_schema` | Generate Schema         | AI-generate a JSON extraction schema from a description.                 |
| `split_document`  | Split Document          | Split a prior extraction into topic-based page ranges.                   |
| `extract_tables`  | Extract Tables          | Extract tables from a prior extraction (HTML or markdown).               |
| `batch_extract`   | Batch Extract           | Extract many documents in one asynchronous batch.                        |
| `run_pipeline`    | Run Extraction Pipeline | Run a saved extraction pipeline on a document.                           |
| `get_job`         | Get Job Status & Result | Poll a long-running job for its status and result.                       |

See [Tools & workflows](/mcp/tools) for each tool's parameters, return shapes, and
end-to-end examples.

## Long-running jobs

Extractions run asynchronously. When a tool can't finish within its inline time budget, it
returns a `processing` stub with a `job_id`, and Claude polls it with **`get_job`** to
fetch the result. You don't manage this yourself, just ask for the result and Claude
chains the calls. See
[asynchronous jobs](/mcp/tools#asynchronous-jobs-and-get_job) for details.

## Parameters and schema shapes

This page keeps the tool list short on purpose. For the full parameter reference, including
the freeform schema-prompt option and the exact shape of `schema` fields, see the
[Pulse API docs](/api-reference/introduction) and the
[structured output guidelines](/api-reference/structured-output-guidelines).

## Next steps

<CardGroup cols={2}>
  <Card title="Tools & workflows" icon="wrench" href="/mcp/tools">
    Full tool reference and end-to-end agent examples.
  </Card>

  <Card title="MCP overview" icon="circle-info" href="/mcp/overview">
    How the Pulse MCP server works and when to use it.
  </Card>
</CardGroup>