This documentation is available as Markdown for AI agents and LLMs. See the full Markdown index or append .md to any documentation URL.

Query EAS Observe metrics, events, and sessions from an AI agent through the Expo MCP Server.

The EAS Observe dashboard and the eas observe commands show metrics, events, and sessions. AI agents can query the same data through the Expo Model Context Protocol (MCP) Server. Connect the server once, and your agent can compare releases, find slow sessions, and read user-defined events for your app. The agent uses the Expo account you sign in with, so it does not need an access token or the EAS CLI.

Prerequisites

Tools

Each tool is read-only and maps to one eas observe command:

Timing values are in seconds. observe_metrics_summary and observe_routes keep Android and iOS results separate, so one platform cannot hide a regression on the other.

Shared inputs

Every tool except observe_session accepts these inputs. The agent fills them in from your prompt:

• appId or appFullName: the project to query. When the agent runs inside the project directory, it reads the project ID from the app config. Otherwise, name the project, such as @owner/my-app.
• platform: one of ANDROID, IOS, IPADOS, MACOS, or TVOS. Without it, observe_versions, observe_metrics_summary, and observe_routes return separate Android and iOS results, and the other tools return results from all platforms.
• days, or start and end: the time range. Defaults to the last 60 days.

observe_session takes the project and a session ID. Session IDs come from observe_metrics and observe_events results.

observe_metrics_summary and observe_metrics accept the same nine metric names as the EAS CLI. See Metric names.

Tools that can return large results paginate them. Each response includes a cursor, and the agent passes it back as nextCursor to continue. Routes paginate per platform, and session timelines paginate metrics and logs separately.

Example prompts

The tools work best in sequence: find the release, compare it, then drill into samples, routes, and one session.

• "Did startup get slower in the latest iOS release?" The agent lists versions with observe_versions, then compares them with observe_metrics_summary.
• "Find the five slowest cold launches this week and tell me what those devices have in common." The agent calls observe_metrics sorted by slowest and reads the device details on each sample.
• "Which routes have the worst time to interactive on Android?" The agent calls observe_routes for the Android platform.
• "Show me the timeline for the session behind that slow launch." The agent takes the session ID from the earlier sample and calls observe_session.
• "How many users completed onboarding since the last update?" The agent discovers the event name with observe_event_names and counts matches with observe_events.

Plans and usage

Plan limits are the same as in the dashboard and the EAS CLI. When your account's plan does not include the requested data, the tool returns an error with an upgrade message. See Pricing for what each plan includes.

Each tool call counts toward your account's Expo MCP Server usage. The server returns the results to the MCP client you connect. See Data privacy for how Expo MCP Server handles project data.

Next steps