Today I am excited to open the Ansight beta: a local bridge that gives coding agents runtime evidence from mobile apps.
Agents are incredible at navigating repositories, proposing fixes, and writing tests. But mobile apps do not live in the repository alone.
They run on simulators and physical devices, inside sandboxed processes with their own lifecycle, permissions, local storage, UI state, and platform APIs.
When an agent can only read the source code, it has to infer too much.
The agent is blind on the other side of the compiler.
Ansight closes that gap with full-fidelity replay, inspection, and control of the running app, so an agent can see what happened, make a change, and verify the result.
Why we built it
Does this sound familiar?
- Find a bug in the running app.
- Try to remember what you tapped.
- Paste screenshots, logs, and rough repro steps into Codex or Claude.
- Wait while the agent guesses at a problem it still cannot see.
That is the loop I wanted to fix.
Ansight captures development and test sessions in high fidelity, so a human can annotate screenshots, capture visual trees and data snapshots, flag FPS drops or memory spikes, and mark the moments that matter. Those notes and artifacts become structured evidence your agent can slice, inspect, and analyse through over 50 MCP tools.
What Ansight does
Ansight connects three pieces:
- The Ansight SDK inside your app captures screenshots, logs, telemetry, navigation state, touch input, local storage, and registered app state.
- Ansight Studio pairs with that app, records sessions, and exposes the runtime surface locally.
- A local MCP bridge lets tools like Codex, Claude, Cursor, Gemini, and Copilot query that evidence and act on it.
That means an agent can see the screen the tester was on, inspect logs and telemetry near that moment, check local storage or SQLite state, and tap through the workflow to verify a fix.
How we use it
We have personally used Ansight with Red-Point: 3D Climbing Guide to:
- Direct Codex and Claude at fiddly UI issues with visual annotations.
- Share captured sessions and extract “Steps to Reproduce”.
- Compare telemetry across runs to validate performance improvements.
- Generate Maestro UI test plans from a recorded session.
Open a public Red-Point captured session.
Once an agent can inspect the same evidence as the developer, prompts get shorter, debugging gets less speculative, and verification stays in the workflow.
Starting with .NET MAUI
The first beta focuses on .NET MAUI apps across iOS, Android, and Mac Catalyst. That is the community and technology stack I know deeply, and it is where I can make the sharpest first version.
Ansight Studio is available today for macOS. Windows builds are coming soon.
Flutter, React Native, native iOS, and native Android support are planned.
Try it in your app
You can ask your agent to install Ansight into a .NET MAUI, Android, iOS, or Mac Catalyst app with the Ansight .NET install skill:
Copyable prompt:
Please use this skill https://www.ansight.ai/skills/dotnet/ansight-install-dotnet.md to install and integrate Ansight into my .NET MAUI, Android, iOS, or Mac Catalyst app. Use the all-in-one Ansight.Maui or Ansight package by default, infer Ansight Studio registration metadata from the app, prefer the Ansight stdio bridge for agent setup, install or recommend the companion inspection skill at https://www.ansight.ai/skills/agents/ansight-app-inspection.md, set AnsightRemoteToolsPolicy to AllowedWithWarnings, inject a global tool guard that enables tools in developer builds and denies all tools in release builds, create ansight-readme.md with QR pairing next steps, and tell me what manual Ansight Studio, agent, or app wiring steps remain.
You can start from the download page or read the docs for installation and MCP setup.
A personal note
Ansight is a personal mission for me.
I have been building tools for the .NET mobile ecosystem since 2014, starting with MFractor, which helped Xamarin developers move faster and reduce pain before I retired it in 2024 alongside Visual Studio for Mac.
Ansight continues that work for the agent-assisted era: less friction, more evidence, and tighter feedback between the app, the developer, and the agent.
Martin and I are incredibly excited to have you along as we close the agentic development loop for mobile. 🤙
