goose

Overview

goose connects to Dash0 in two directions. Our remote MCP server lets goose read your telemetry data, navigate and analyze it, while leveraging the same capabilities that human users have access to in the UI. Separately, goose emits OpenTelemetry data about its own sessions, which you can send to Dash0 to see what the agent did and what it cost.

Set up whichever direction you need. They are independent.

Prerequisites

• goose installed. Version 1.46.0 or later is required for the GenAI semantic convention attributes described below.
• A Dash0 auth token.
• Your MCP and OTLP endpoint URLs, both visible under Settings → Endpoints in Dash0.

Connect goose to the Dash0 MCP server

The Dash0 MCP server supports two authentication methods:

• OAuth 2.0 (recommended) — browser-based consent on first connection, no static token to manage. Access tokens refresh automatically every 15 minutes.
• Auth token (Bearer) — best for CI/CD pipelines, scripts, and headless environments where browser-based consent is impractical.

For more details on token lifecycle, audit logs, and revocation, see the Dash0 MCP Server integration.

Note that goose uses uri where most other MCP clients use url, and streamable_http with an underscore.

Option 1: OAuth 2.0 (Recommended)

1. Add the extension to ~/.config/goose/config.yaml:

   You can also add it interactively. Run goose configure, choose Add Extension → Remote Extension (Streamable HTTP), enter {{endpoint_mcp}} as the URL, and skip the headers.

2. Start a session. On first connection, goose registers itself with Dash0 through dynamic client registration and opens a browser window where you log in and grant consent. You can review and revoke connected applications under User Settings → Applications in Dash0.

Option 2: Auth Token

Before configuring the Dash0 MCP server with a Bearer token, create an auth token with the correct permissions:

1. Go to your organization's Auth Tokens settings page
2. Create a new auth token with one of the following permission levels:
   • Full access (*) - Recommended for full MCP functionality
   • Read-only (*:read) - For read-only access to your telemetry data

Add the extension to ~/.config/goose/config.yaml:

Alternatively, run goose configure, choose Add Extension → Remote Extension (Streamable HTTP), and enter:

• Name: dash0
• URI: {{endpoint_mcp}}
• Header: Authorization: Bearer {{token}}

Verify

Run goose info -v to confirm the file still parses and the extension is listed. Then start a session and ask goose what it can reach:

Tools such as getServiceCatalog, getSpans, promql, and listDashboards confirm the connection.

Send goose telemetry to Dash0

goose links the OpenTelemetry SDK and reads the standard OTEL_* environment variables at startup, initializing traces, metrics, and logs if it finds an endpoint. No plugin or code change is needed.

Set the data export endpoint and authorization token:

goose appends /v1/traces, /v1/metrics, and /v1/logs itself, so leave the path off the endpoint.

You can specify which dataset to send data to by adding the Dash0-Dataset header. For example:

For more information, see the dataset documentation.

What goose reports

• Traces: one trace per session, with gen_ai.operation.name set to invoke_agent, chat, or execute_tool. Inference spans carry gen_ai.request.model, gen_ai.provider.name, and per-call token usage. Tool calls carry gen_ai.tool.name and gen_ai.tool.call.id.
• Metrics: goose.* counters covering sessions, tool calls, recipe runs, CLI commands, and the prompt injection scanner. The exact set changes between goose releases, so browse the goose. prefix in the Metrics Explorer to see what your version emits.
• Logs: session lifecycle records such as Tool call started and Session completed.

Verify

1. Run any prompt, for example goose run -t "Say hello".
2. Within a few seconds, a trace named reply appears in Dash0 under the goose service.

References

• goose documentation and repository
• Using extensions in goose
• OpenTelemetry GenAI semantic conventions