Integration Tools

Access external services through MCP connectors like Linear, Slack, and Datadog.

When you connect MCP services to Sazabi, Chat gains access to tools from those services. This extends the assistant beyond observability data into your broader development workflow, allowing it to create issues, send messages, and take actions across your toolchain.

How integration tools work

MCP connectors expose tools that the assistant can call during conversations. Each connected service provides a different set of capabilities:

Service	Example tools
Linear	Create issues, update status, list projects
Slack	Send messages, read channel history, add reactions
Datadog	Query metrics, create monitors, manage incidents
incident.io	Create incidents, access runbooks, update status

When you ask the assistant to perform an action that requires an external service, it selects the appropriate tool and calls it with the relevant parameters.

Integration tools only appear after you connect the corresponding MCP connector. Go to Settings -> MCP Connectors to connect services.

Available integration tools

The assistant can use tools from any MCP connector you have connected. The specific tools depend on what each service's MCP server provides.

Pre-configured connectors

Sazabi provides pre-configured connectors for these services:

Slack: Send messages, access channels, add reactions
Linear: Create and manage issues, projects, and cycles
Datadog: Query metrics and create monitors
incident.io: Manage incidents and runbooks
Supabase: Query databases and manage auth
Sentry: Access error data and manage issues

Custom MCP servers

You can also connect custom MCP servers to give the assistant access to your internal tools. Any MCP-compatible server can be connected, exposing its tools to the assistant.

Authentication and permissions

OAuth-based connectors

Most pre-configured connectors use OAuth authentication:

You authorize Sazabi to access the service on your behalf.
Sazabi stores the credentials securely.
The assistant uses your permissions when calling tools.

This means the assistant can only do what you can do in the connected service. If you cannot create issues in a Linear team, the assistant cannot either.

Custom server authentication

Custom MCP servers use header-based authentication. You provide the necessary headers (API keys, tokens, etc.) when configuring the connection. These credentials are encrypted at rest.

Anyone with admin access to your Sazabi organization can configure MCP connectors and see tool outputs. Consider this when connecting services with sensitive data.

Organization-scoped access

MCP connectors are configured at the organization level. When you connect a service, all members of your organization can use that connection through the assistant. Individual members do not need to authorize separately.

Using integration tools

Natural language requests

Ask the assistant to use integration tools in natural language:

Create a Linear issue for this error with the stack trace
Post a summary of this investigation to #incidents on Slack
Check if there are any related incidents in incident.io

The assistant determines which tools to use based on your request and the available connectors.

Tool visibility

When the assistant uses an integration tool, you see:

Tool name: Which tool was called (e.g., linear_create_issue)
Parameters: The inputs the assistant provided
Results: What the external service returned

You can expand any tool call to see full details, including the exact data sent to and received from the external service.

Automatic suggestions

During investigations, the assistant may suggest using integration tools when appropriate. For example:

After finding a bug, it may offer to create a Linear issue
After completing an investigation, it may offer to post a summary to Slack
When detecting a severe problem, it may suggest creating an incident

You can approve or modify these suggestions before the action is taken.

Example workflows

Bug investigation to issue tracking

You ask the assistant about elevated error rates.

The assistant searches logs and identifies a null pointer exception in the payment service.

The assistant offers to create a Linear issue with the error details, affected endpoints, and sample stack trace.

You approve, and the assistant creates the issue with a link back to the investigation thread.

Investigation summary to Slack

You complete an investigation into a database connection issue.

You ask the assistant to summarize and post to the #backend channel.

The assistant formats the key findings and uses the Slack connector to post the summary.

Team members in Slack see the summary with a link to the full investigation.

Incident coordination

The assistant detects a critical error pattern during monitoring.

It creates an incident in incident.io with severity and initial details.

It posts an alert to the #incidents Slack channel with the incident link.

It creates a Linear issue to track the fix.

Custom MCP server setup

You can connect any MCP-compatible server to extend the assistant with custom tools.

Requirements

Your MCP server must:

Be accessible over HTTPS
Implement the MCP protocol specification
Return tool definitions that describe available capabilities
Respond to tool calls with valid MCP responses

Connecting a custom server

Go to Settings -> MCP Connectors in the Sazabi dashboard.

Click Add Custom Server.

Enter a name, the server URL, and any required authentication headers.

Click Connect to test the connection and discover available tools.

Example: Internal deployment tools

If your team has an MCP server that exposes deployment and infrastructure tools:

Name: Internal DevOps
URL: https://mcp.internal.example.com/mcp
Headers:
  X-API-Key: your-api-key

Once connected, you can ask the assistant:

Check the deployment status of the API service
List recent deployments to production
Show the health status of our Kubernetes pods

For detailed setup instructions, see the Custom MCP Server documentation.

Verify the connector is connected in Settings -> MCP Connectors.
Check that tools are listed under the connector.
Try disconnecting and reconnecting to refresh the tool list.

Tool calls failing

If tool calls return errors:

Check that the OAuth authorization has not expired.
Verify you have the necessary permissions in the external service.
Review the error message in the tool call output for details.