Integrations
Integrations connect Autoheal to your observability stack, allowing the AI agent to query metrics, logs, traces, and alerts during incident investigations.
Autoheal uses the Model Context Protocol (MCP) to communicate with integrations. MCP is an open standard developed by Anthropic that defines how AI agents connect to external tools and data sources. It provides a standardized way for the AI agent to interact with different tools while maintaining security and access control.
With MCP, integrations expose specific capabilities (like "query metrics" or "search logs") that the AI agent can invoke during investigations. This means the agent can only perform actions that your integration explicitly permits—there's no hidden access or unexpected behavior.
Available Integrations
| Integration | Category | Description |
|---|---|---|
| Datadog | Monitoring | Metrics, APM, logs, incidents |
| Grafana | Monitoring | Dashboards, metrics, alerts |
| Sentry | Error Tracking | Errors, exceptions, performance |
| GitHub | Source Control | Code, PRs, deployments, Actions |
| GitLab | Source Control | Code, MRs, pipelines |
| Confluence | Documentation | Wiki pages, spaces |
| Notion | Documentation | Pages, databases |
| Honeycomb | Observability | Traces, events, queries |
| Elasticsearch | Logs | Log search, aggregations |
| ClickHouse | Analytics | SQL analytics queries |
| Databricks | Data Store | SQL queries, job monitoring, clusters |
| MS SQL Server | Data Store | SQL queries, logs, metrics |
| Neo4j | Data Store | Graph database, knowledge graphs |
| PagerDuty | Incident Management | Incidents, on-call, alerts |
| PostgreSQL | Data Store | SQL queries, logs, metrics |
| Pylon | Customer Support | Ticket investigation, internal notes |
| ReadMe Docs | Documentation | Search and read documentation |
| Slack | Chat | @mention investigations, alerts |
| Custom MCP | Custom | Your own MCP servers |
General Setup Process
All integrations follow a similar setup process:
From the Autoheal sidebar, click Integrations.
Click on the integration you want to add from the available list.
Give your integration a descriptive name (e.g., "Production Datadog", "Staging Grafana").
Provide the required credentials. These typically include:
- API Key or Token
- Base URL (for self-hosted instances)
- Additional configuration options
Click Test Connection to verify the integration is configured correctly.
Click Save to activate the integration.
Security & Credentials
Credential Storage
- Credentials are stored in AWS Secrets Manager, isolated per tenant
- Access is scoped to your organization only
- Credentials are never logged or included in error messages
Best Practices
Use Service Accounts
Create dedicated service accounts or API keys for Autoheal rather than using personal credentials. This makes it easier to:
- Audit access
- Rotate credentials
- Limit permissions
Principle of Least Privilege
Grant only the permissions Autoheal needs:
- Read-only access is sufficient for most integrations
- Avoid granting write or admin permissions unless necessary
Regular Rotation
Rotate API keys periodically:
- Set calendar reminders for key rotation
- Update credentials in Autoheal immediately after rotation
- Test the connection after updating
Multiple Instances
You can connect multiple instances of the same integration type. Common use cases:
- Production vs Staging - Separate Datadog accounts for different environments
- Regional Instances - Different Grafana instances per region
- Team-Specific - Dedicated monitoring for different teams
Each instance will appear as a separate option when the AI agent investigates.
Troubleshooting
Connection Failed
Invalid API Key
- Verify the API key is correct and hasn't been rotated
- Check that the key has the required permissions
- Ensure the key hasn't expired
Network/Firewall Issues
- Verify Autoheal can reach your integration's API endpoint
- Check if your firewall allows inbound connections from Autoheal IPs
- For self-hosted tools, ensure the URL is accessible
Permission Denied
- Verify the API key has read permissions
- Check organization-level access controls
- Some integrations require specific scopes or roles
Integration Not Appearing in Investigations
- Ensure the integration status shows as Connected
- Verify the integration is enabled (not disabled)
- Check that your user has permission to access the integration