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

# MCP Introduction

> Connect GC AI to ChatGPT, Claude, Cursor, and other MCP clients so an assistant can work in your knowledge base on your behalf.

The GC AI MCP server lets an AI assistant work directly in your GC AI account. Connect it once in a host like ChatGPT, Claude, or Cursor, and the assistant can search your files, run playbooks, ask legal questions grounded in your knowledge base, and organize projects and skills, all as you.

[MCP](https://modelcontextprotocol.io) (the Model Context Protocol) is the open standard those hosts use to call external tools. The GC AI server implements it over a single authenticated endpoint, so any MCP-compatible client can use it.

If you are building a backend integration or script rather than connecting an assistant, use the [REST API](/api-reference/introduction) instead. The two cover the same ground; MCP is the right fit when an AI agent is the one doing the work.

## What you can do

The server exposes 20 tools, grouped by what they touch:

* **Ask GC AI.** Ask a legal question or run a task grounded in your files, playbooks, and skills, then save the resulting chat into your history.
* **Files.** Upload documents, poll them until they finish processing, and search across everything you can access.
* **Projects.** Create and update projects, and attach or detach files to group work around a matter.
* **Playbooks.** List, create, and update review playbooks, and run one against uploaded files to get structured check results.
* **Skills.** List, create, and update reusable instructions the assistant can apply on demand.

Every tool page under this section lists its exact inputs, outputs, and behavior. The [Surface map](/api-reference/mcp/surface-map) shows all 20 tools at a glance; see [Connecting](/api-reference/mcp/connecting) to set the server up in your host.

## How requests are scoped

Requests run as you, the signed-in user, not as a shared service account. The assistant sees exactly what you can see in GC AI: your files, the projects and playbooks shared with you, and your organization's open resources. Nothing it does reaches beyond your own access.

Usage is billed to your organization, the same as the rest of GC AI.

## Two behaviors worth knowing

A few tools work differently from a plain request-and-response call. Good MCP hosts handle both for you, but it helps to know what is happening.

<AccordionGroup>
  <Accordion title="Long-running work returns a job to poll">
    `ask_gcai` and `run_playbook` can take a while, so they return a job with a `status` and a `job_id` rather than a finished answer. The assistant polls the matching status tool (`ask_gcai_status`, `run_playbook_status`) until the job succeeds and the result is ready. This is expected, and the assistant does it on its own.
  </Accordion>

  <Accordion title="File uploads process in place">
    An uploaded file is stored and then processed (text extraction and embeddings) before it is ready to use. After an upload, the assistant polls `get_files` with the file's `id` until `status` is `ready`. Two upload paths exist: `upload_file` for a file your host has already attached, and `start_file_upload` plus `finish_file_upload` for a file that lives only in the assistant's own sandbox.
  </Accordion>
</AccordionGroup>

## Next steps

<CardGroup cols={2}>
  <Card title="Connect a host" icon="plug" href="/api-reference/mcp/connecting">
    Add the GC AI server to ChatGPT, Claude, or Cursor.
  </Card>

  <Card title="Browse the tools" icon="wrench" href="/api-reference/mcp/tools/ask-gcai">
    See every tool's inputs, outputs, and behavior.
  </Card>
</CardGroup>