MCP connectors let [Global Agent](/docs/amplitude-ai/global-agent-overview) reach into the tools where your team plans, ships, and talks, so it can ground answers in product specs, tickets, and conversations alongside your Amplitude data and [page context](/docs/amplitude-ai/adding-context). Use connectors to ask cross-source questions like "Which retention drop correlates with the Jira incidents last sprint?" and to take action in those tools without leaving Amplitude. This is useful for product managers connecting analytics to ongoing work, and for technical teams who want the agent to read and update tickets, docs, and repositories.

## Supported connectors

Global Agent supports the following MCP connectors:

| Connector                        | What Global Agent can do                                                              |
| -------------------------------- | ------------------------------------------------------------------------------------- |
| **Atlassian** (Jira, Confluence) | Read Jira tickets and Confluence pages, create and update issues, and comment.        |
| **GitHub**                       | Read pull requests, issues, and code, open issues, and comment on pull requests.      |
| **Granola**                      | Read meeting notes and transcripts to bring discussion context into the conversation. |
| **Linear**                       | Read issues and projects, and create and update issues.                               |
| **Notion**                       | Read pages and databases, and create and update pages.                                |
| **Sentry**                       | Read error and issue details to connect production problems with analytics.           |
| **Slack**                        | Search and read channels and threads, and post messages.                              |

Organization Admins control which of these connectors org members can authorize from the **Connectors** tab in [AI Settings and Controls](/docs/amplitude-ai/ai-controls). For more detail, refer to [Manage available connectors](#manage-available-connectors).

## How MCP connectors work

Global Agent uses the [Model Context Protocol (MCP)](https://modelcontextprotocol.io) to connect with external services. Each connector exposes a set of tools that Global Agent can call during a conversation. Tools fall into two categories:

- **Read tools** pull information into the conversation, like fetching a Jira ticket, reading a Notion page, or searching a Slack channel.
- **Write tools** take action in the connected service, like creating a ticket, updating a doc, or posting a message.

You authorize each connector through [OAuth](https://oauth.net/2/), so Global Agent acts on your behalf and respects the same permissions you have in the connected tool. Once connected, that service's tools stay available to Global Agent in every conversation until you disconnect the connector.

## Use cases

- **Correlate analytics with product work**: Ask Global Agent to compare a retention drop against Jira incidents or Linear bugs from the same sprint.
- **Ground analysis in product specs**: Pull a Notion or Confluence spec into the conversation, then ask Global Agent to find the matching funnel or cohort.
- **Close the loop on insights**: Turn an analysis into a Jira ticket, Linear issue, or Slack post without leaving Global Agent.
- **Summarize cross-tool context**: Ask Global Agent to combine a GitHub PR description, the related Linear ticket, and an Amplitude chart into a single summary.

### Take action with write tools

Connectors support both read and write tools, so Global Agent can do more than retrieve context. You can ask Global Agent to:

- Create a Jira or Linear ticket from an insight.
- Update a Confluence or Notion page with findings from an analysis.
- Post a summary to a Slack channel.
- Open a GitHub issue or comment on a pull request.

Global Agent confirms write actions before it runs them, so you stay in control of what changes in the connected tool.

## Prerequisites

- Global Agent enabled for your organization. Organization Admins manage availability from [AI Settings and Controls](/docs/amplitude-ai/ai-controls).
- An account in the external service you want to connect, with permission to authorize third-party apps.

## Set up a connector

1. Open Global Agent.
2. Click the **Select Mode** icon and then select **Connectors > Add Connectors**.
3. Select the connector you want to enable, then click **Connect**.
4. Complete the OAuth flow in the popup window to authorize Global Agent.
5. Return to Global Agent. The connector now appears as available in the Connectors panel.

Repeat for each connector you want to use. Connectors only need authorization one time for each account.

## Manage available connectors

By default, every supported MCP connector is on your organization's allowlist, so any org member can connect any service. Organization Admins manage the allowlist from [AI Settings and Controls](/docs/amplitude-ai/ai-controls) to restrict which connectors org members can authorize.

1. Go to _Settings > AI Controls_.
2. Open the **Connectors** tab to view the connectors available to your org members.
3. To remove a connector from the allowlist, open the more-options menu (**...**) on its row and remove it. Org members can no longer connect a removed service from Global Agent.
4. To add a connector back, click **Browse MCP** and select it from the catalog.

## Frequently asked questions

{% accordion title="Which connectors does Global Agent support?" %}
Global Agent supports seven MCP connectors: Atlassian (Jira and Confluence), GitHub, Granola, Linear, Notion, Sentry, and Slack. For what each connector can do, refer to [Supported connectors](#supported-connectors). Your Organization Admin controls which of these connectors you can authorize.
{% /accordion %}

{% accordion title="Do I need to authorize a connector every time I use it?" %}
No. You authorize each connector one time for each account. After you complete the OAuth flow, that service's tools stay available to Global Agent in every conversation until you disconnect the connector.
{% /accordion %}

{% accordion title="Can Global Agent make changes in my connected tools?" %}
Yes. Connectors support both read tools that pull information into the conversation and write tools that take action, such as creating a ticket, updating a page, or posting a message. Global Agent confirms each write action before it runs, so you stay in control of what changes in the connected tool.
{% /accordion %}

{% accordion title="What can Global Agent access in a connected service?" %}
Global Agent acts on your behalf and respects the same permissions you have in the connected tool. It can't read or change anything you can't already access yourself.
{% /accordion %}

{% accordion title="Why don't I see a connector I expected?" %}
Organization Admins manage which connectors org members can authorize from the **Connectors** tab in [AI Settings and Controls](/docs/amplitude-ai/ai-controls). If a connector you expected isn't available, your admin may have removed it from the allowlist. Ask your Organization Admin to add it back.
{% /accordion %}

{% accordion title="Is my connected-tool data stored or used to train models?" %}
Amplitude doesn't store your data with third-party AI providers, and Amplitude and its partners never use your data for model training. For more detail, refer to [Privacy and security](/docs/amplitude-ai/privacy-and-security).
{% /accordion %}