@agentrysh/mcp

The canonical agentry entry point. Drop it into Claude Code, Cursor or any MCP-compatible AI and your assistant can read your app's errors, analytics and deploys directly — no dashboard, no docs, no query language to learn.

Install (Claude Code)

claude mcp add agentry -- npx -y @agentrysh/mcp

Then in your session:

set me up with agentry

The agent walks you through a device-flow login (you sign in at agentry.sh/cli — GitHub, Google, or magic-link email, your pick), mints an API key, creates an agentry project, and hands you a paste-ready snippet for your app.

What you get

Once installed, your AI assistant can:

• Read errors — agentry_list_cases, agentry_get_case (with stack, breadcrumbs, deploy attribution)
• Run named queries — agentry_run_recipe (14 canonical queries: DAU, funnels, retention, top errors, deploy health, weekly digest)
• Ask anything in HogQL — agentry_analytics_query
• Send events — agentry_capture_test_event, agentry_record_deploy, agentry_track_test_event
• Manage suppressions, webhooks, alerts — full CRUD on the QoL layer

What agentry is

A small, agent-first observability backend. Three HTTP endpoints, one DSN:

• POST /v1/logs/{project_id}/
• POST /v1/analytics/{project_id}/
• POST /v1/deploys/{project_id}/

Send JSON. Your AI does the rest.

See agentry.sh for the full pitch.

License

MIT