Connect Your First Integration
A step-by-step tutorial to connect external services so your agents and workflows can act on them.
This tutorial walks you through connecting your first external service to Fabric AI. In about 10 minutes, you'll have an integration ready to use in AI chat and workflows.
What You'll Accomplish
By the end of this tutorial, you'll have:
- Connected an external service (GitHub, Slack, or Notion)
- Tested the connection
- Used the integration in an AI conversation
- Understood how integrations work
Prerequisites
Before starting, make sure you have:
- ✅ A Fabric AI account
- ✅ An account with at least one external service (GitHub, Slack, Linear, Notion, etc.)
How Integrations Work
Fabric connects to external systems so your agents and workflows can take actions on them at runtime — search a workspace, read a page, create an issue, post a message. You manage every connection in one place: Settings → Integrations.
A connection is used in two ways:
| Used by | What it powers | Example |
|---|---|---|
| MCP Servers | AI-driven tool calls in chat and the orchestrator | "Create a GitHub issue", "post to Slack" |
| Workflow Integrations | Action nodes in the visual workflow builder | Scheduled reports, triggered actions |
Both share the same connected credentials — connect a system once and it's available to agents and workflows alike.
Each provider exposes a set of runtime actions in two flavors: knowledge actions that read or search (e.g. "Search Notion", "Get Page Content") and write actions that change things (e.g. "Create Page"). A provider that offers both is labeled Hybrid.
Fabric runs these as live runtime actions — there's no background sync or separate "data connection" to set up. When an agent needs information from a connected system, it calls a knowledge action on demand. The one structured exception is backlog sync to a project-management tool — see Project Management Sync.
Step 1: Set Up an MCP Server
MCP servers let AI agents interact with external tools during conversations.
Navigate to MCP Settings
Go to Settings → MCP Servers in the left sidebar.
Browse the Registry
Click Add Server to see all available MCP servers organized by category:
- Development — GitHub, GitLab, Atlassian (Jira & Confluence), Azure DevOps, Fizzy
- Communication — Slack
- Knowledge & Docs — Notion, Google Drive
- Diagramming — Excalidraw
Select and Connect
Pick a server — we'll use GitHub as an example:
Option A: OAuth (Recommended)
- Click Connect with OAuth
- Authorize Fabric AI in the GitHub popup
- Done — tokens are stored and auto-refreshed
Option B: API Key
- Generate a Personal Access Token on GitHub
- Paste it into the API key field
- Click Save
Test the Connection
Click Test Connection. You should see a success message with latency info.
Try It in Chat
Open AI Chat and try a command:
"List my open GitHub issues"The AI discovers the GitHub MCP tools and uses them to fetch your issues.
Step 2: Set Up a Workflow Integration
Workflow integrations power the visual workflow builder nodes.
Open the Workflow Builder
Navigate to Workflows and open or create a workflow.
Add an Integration Node
Click + to add a node, then browse the Integrations category. Select the service you want (e.g., Slack, GitHub, Linear).
Configure Credentials
Click Configure Integration on the node:
- Enter your API key or token
- Click Test Connection to verify
- Credentials are encrypted and stored securely
Configure the Action
Set up what the node does:
- GitHub: Create Issue → set owner, repo, title
- Slack: Send Message → set channel, message
- Linear: Create Issue → set team, title, priority
Integration Scope
Personal vs Organization
| Scope | Visibility | Use Case |
|---|---|---|
| Personal | Only you | Your personal GitHub, personal Drive |
| Organization | All org members | Company Jira, shared Slack workspace |
Choose the scope when connecting. Organization integrations are managed by org admins.
Multi-Tenant Isolation
- Personal connections are never visible in organization contexts
- Organization connections are never visible to other organizations
- Each user can have their own connections alongside org-shared ones
Quick Reference: Supported Integrations
Type is the kind of runtime actions a provider offers: Knowledge (read/search), Actions (write), or Hybrid (both).
| Service | Type | Set up under |
|---|---|---|
| GitHub | Hybrid | Settings → Integrations |
| GitLab | Hybrid | Settings → Integrations |
| Slack | Hybrid | Settings → Integrations |
| Microsoft Teams | Hybrid | Settings → Integrations |
| Jira | Hybrid | Settings → Integrations |
| Confluence | Knowledge | Settings → Integrations |
| Notion | Hybrid | Settings → Integrations |
| Google Drive | Knowledge | Settings → Integrations |
| Linear | Hybrid | Settings → Integrations |
| Azure DevOps | Hybrid | Settings → MCP Registry |
| Excalidraw | Actions | Settings → MCP Registry |
| Fizzy | Hybrid | Settings → MCP Registry |
Microsoft 365 also appears under Settings → Integrations, but its runtime actions aren't available yet ("Actions setup not yet available"). Use Microsoft Teams for Microsoft Graph actions today.
Troubleshooting
"Connection failed" during setup
- Double-check your API key or token
- Ensure the token has the required permissions/scopes
- Try disconnecting and reconnecting for OAuth-based services
Tools not appearing in AI Chat
- Verify the MCP server is Enabled in settings
- Check that the server's health status is Healthy
- Try refreshing the tool list in the chat settings
An action isn't available for a provider
- Open the provider under Settings → Integrations and check Available actions — some providers expose only knowledge actions, others only write actions
- Make sure the connection shows Connected and the required scopes were granted during authorization