> For the complete documentation index, see [llms.txt](https://notara-1.gitbook.io/notara-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://notara-1.gitbook.io/notara-docs/mcp.md).

# MCP Setup

Notara exposes a **Model Context Protocol (MCP)** endpoint that lets any MCP-compatible agent — Claude Code, Cursor, or your own code — access your workspace's governed tools and context.

## What Is the MCP Endpoint?

The MCP endpoint is a server-sent events (SSE) endpoint that speaks the [Model Context Protocol](https://modelcontextprotocol.io) standard. When you connect an external agent to it, that agent gains access to:

* Your Notara context libraries (as MCP resources)
* Your connected tool connectors (as MCP tools)
* Your team's skills (as MCP prompts)

The key distinction from a generic API: **the same governance applies**. Every tool call through MCP is subject to the same capability defaults, tenant permission overrides, and approval flows that govern your Slack agent. An MCP caller can never do more than an authorized team member asking through Slack.

> **On Free:** MCP works on Free. Your AI client sees the workspace's own skills (up to 5) as prompts plus the context tools for searching and reading your libraries. Connected tools are paid. Free includes 1,000 context requests per month per workspace. See [Free plan](/notara-docs/billing/free-plan.md).

## Why Connect External Agents?

Your team may already use Claude Code for code generation, Cursor for coding, or custom agents for automation. Rather than maintaining separate context in each tool, connecting them to Notara means:

* **Shared context**: your company handbook, runbooks, and API docs are available to your coding agent — no copy-pasting.
* **Governed tools**: Linear, GitHub, Stripe, etc. are accessible from your IDE with the same approval chains your team already configured.
* **Audit trail**: every tool call from an external agent is logged in Notara's audit log, alongside Slack-originated calls.
* **Group scoping**: MCP endpoints are scoped to a group — your engineering group's endpoint has different tools and context than your growth group's.

## Endpoint URL Format

```
https://app.notara.ai/mcp/g/<workspace-id>/<group-slug>
```

For example:

```
https://app.notara.ai/mcp/g/acme-corp/engineering
```

You'll find your exact endpoint URL in the dashboard under **Connect → MCP Endpoints**, next to the endpoint you want to use.

## Authentication

MCP endpoints use Bearer token authentication. Tokens are scoped to a single group endpoint and can be revoked individually.

To generate a token:

1. Go to **Connect → MCP Endpoints** in the dashboard.
2. Click **New Endpoint** (or open an existing one).
3. Click **Generate Token**.
4. Copy the token — it's only shown once. Store it securely.

Tokens do not expire by default but can be revoked at any time from the same panel.

## Groups and Scoping

Each MCP endpoint is tied to a specific **group** within your workspace. Groups are sub-teams that have their own:

* Context collections
* Skills
* Connector access
* Tool permission overrides

This means your engineering group's MCP endpoint exposes different tools and context than your marketing group's. When creating an endpoint, choose the group whose context and tools are relevant to the agent you're connecting.

## What Gets Exposed

| MCP Primitive | What It Contains                                                   |
| ------------- | ------------------------------------------------------------------ |
| **Resources** | Documents from the group's context libraries (spine and on-demand) |
| **Tools**     | Connector tools the group has access to, filtered by permission    |
| **Prompts**   | Skills in the group's skill library                                |

Tools are built from the persisted cache — the tool surface doesn't do a live `tools/list` on every request, so it's fast and consistent. The cache is refreshed when you add or refresh a connector.

## Connect Your Agent

* [Claude Code](/notara-docs/mcp/claude-code.md) — add to `~/.claude/settings.json`
* [Cursor](/notara-docs/mcp/cursor.md) — add via Cursor's MCP settings panel
* [Other MCP clients](/notara-docs/mcp/other-clients.md) — general connection guide