Hi everyone,

We've teased it in the last two newsletters, and many of you have been using the beta for months: the Avo MCP is officially launched. Your AI assistants can now read and write your tracking plan, under the same governance your team already runs in Avo.

Also in this newsletter:

• Library Interface codegen lands on TypeScript
• Faster exports, variant overrides, Inspector and Figma updates
• A teaser of what is coming up

🤖 The Avo MCP, officially launched

Connect Claude, Cursor, ChatGPT, Codex, Gemini, or any MCP-compatible client to your Avo workspace, and your tracking plan becomes something your AI can work with directly. Day to day, that looks like:

• Ask your plan questions. "Do we already track starting a checkout?" finds Purchase Started by meaning, with its full schema, even when the name doesn't match the words.
• Design on a branch, under your rules. Hand the agent a PRD or a Figma and it drafts tracking onto an Avo branch, following your naming conventions and audit rules and reusing your existing events and properties instead of inventing duplicates.
• Implement from your branches. Implementation guides and per-source code snippets come straight from the branch diff, so your developers and their coding agents ship exactly what was designed.
• Query with your definitions. Because Avo holds your metric definitions, agents compute conversion in your warehouse or analytics platform the way your team defined it.

If you tried the beta early, a lot has landed since: full write coverage in save_items (tags, categories, owners and stakeholders, name mappings, variant overrides, metrics, archiving and restoring), branch review status and pull-main for agents, the same Editor/Admin role gates as the UI, and a big performance upgrade on large workspaces.

The design method ships with it too: our open-source agent skills install as a Claude Code plugin (one install, MCP server included) and teach the agent to design tracking the way your plan expects.

And the governance you rely on doesn't move: every write lands on a branch, never on main. The agent proposes; your team decides what merges, in the same review flow you use today.

We wrote it up in two posts worth your time. The launch post runs an experiment where the same agent designed the same feature twice, once blind and once reading the live tracking plan: same model, different harness, very different result. And the companion piece, Govern your data. Your agents depend on it., lays out the three beliefs about data in an agent-first world behind all of it, and why your tracking plan is the live source of truth your agents should reach for first.

Read the docs → Avo MCP overview

That's the MCP. Here's what else has shipped since the last newsletter.

🏗️ Codegen: Library Interface lands on TypeScript

Last issue we said the Library Interface mode was coming to TypeScript. It's here, including custom destinations, and it combines with the one-file-per-event mode, so you get tree-shakeable modular events and a stable shared library file in the same setup.

Two more modernization updates for Android: generated networking code now uses Kotlin coroutines instead of the deprecated AsyncTask, and when Inspector is enabled, generated code calls it via direct imports instead of reflection. Cleaner builds, friendlier to ProGuard/R8.

Read the docs → Library Interface mode

🧰 Smaller wins

• Faster exports. CSV exports are about 5x faster and JSON Schema exports about 2x faster, across all exporters.
• Snappier branch diffs. Reviewing large branches got noticeably faster after a big main-thread performance pass.
• Per-variant destination overrides. Event variants can now override which destinations they're sent to.
• Inspector issues keep their history. The issues view now considers a 30-day activity window instead of 2 days, so issues no longer vanish after a couple of quiet days.
• Figma import polish. Clearer errors with retry guidance when Figma rate-limits, seat-type recommendations when connecting, and a notice when your monthly import quota resets.

🔮 What's next

The MCP is one step on a path we're already building: the agent-first workflow, where agents design, implement, verify and query analytics end to end, and pull a human in when they're not confident or when your governance rules require it. Next up: an agent inside the Avo UI, and journeys and review workflows through the MCP.

One more, in beta right now: custom fields and PII classification brings governance metadata and data classification into the tracking plan. Enterprise workspaces can request beta access, with a full announcement at GA. If this is top of mind for you, reply and we'll get you in.

As always, we'd love your feedback. If the MCP changes how your team designs tracking, hit reply and tell us about it.

Happy prompting 🎉