--- title: "bitdrift's MCP server brings always-on agentic investigations to your whole team" metaTitle: "bitdrift's MCP server for always-on agentic investigations" slug: "mcp" blurb: "Anyone can now query your mobile fleet from anywhere, in their AI agent of choice, via the bitdrift MCP server." metaDescription: "Anyone can now query your mobile fleet from anywhere, in their AI agent of choice, via the bitdrift MCP server." cover: url: "/assets/posts/mcp/feature-mcp-launch-desktop@1x.webp" alt: "bitdrift MCP server launch" socialThumbnail: url: "/assets/posts/mcp/feature-mcp-launch-desktop@1x.webp" alt: "bitdrift MCP server launch" author: - "miguel" tags: - "feature" - "ai" - "news" publishedDate: "2026-10-28T12:00:00.000Z" modifiedDate: "2026-10-28T12:00:00.000Z" --- ## Mobile observability for anyone, from anywhere, using any agent It’s 8am on a Tuesday. You, a product manager for a mobile app, are on the train. Yesterday your team shipped a redesign of the app’s home screen. As you’re wondering how it’s landing, your phone buzzes. It’s your ChatGPT dot: > “Heads up: app terminations spiked after yesterday’s release. Want me to dig in?” You ask in reply, “Can you show me a chart of terminations by type for the last 90 days?” The agent pulls the data from bitdrift and draws an interactive chart right in the chat. The spike is hard to miss, and you tap into it to see which types of exits are driving it and which sessions were hit. Then you follow up: “For all the force quits, can you show me the paths leading up to them?” A minute later, you’re looking at every screen journey that ended with a user quitting the app, and most of them run through the new home tab. You didn’t install anything, or open bitdrift, or touch a terminal, or ask an engineer to pull a report. You didn’t even ask the first question. The AI agent you already use was keeping an eye on the release, using bitdrift. Today, anyone on your team can query your mobile fleet from anywhere, in their AI agent of choice, via [the bitdrift MCP](https://docs.bitdrift.io/ai/mcp). ## Agentic investigations, but with full context Spotting a spike in terminations is a good start, but bitdrift goes much further. A termination count tells you that users left. Because bitdrift captures full sessions from every device, not a sample, your agent can show you what the user was looking at right before quitting the app. You can ask: > “Can you share the last screen capture you have for a user that force quit?” The agent finds the most recent session that ended in a force quit and pulls its last screen capture. A 3D render of that screen’s layout shows up right in the chat, in a rich and interactive UI. You can rotate it and zoom in to see exactly what they were looking at, and one click opens the full session in bitdrift. ## So, why an MCP server? Earlier this year we launched [bitdrift AI](https://blog.bitdrift.io/post/introducing-bitdrift-ai), the first agentic mobile observability platform. Getting there took three steps: 1. A [Public API](https://blog.bitdrift.io/post/public-api) and [docs](https://docs.bitdrift.io/api-guide/), so everything in bitdrift could be reached programmatically. 2. The [bd CLI](https://docs.bitdrift.io/cli/quickstart), so people could work with that API from a terminal. 3. [bd skills](https://github.com/bitdriftlabs/bd-skills/), which teach an agent how to use bitdrift: reading the docs, instrumenting an app, and deploying dashboards to monitor critical user journeys. With these building blocks in place, agents could collect new telemetry from any device without an app release and work through issues. Customers put bitdrift AI to work in all kinds of interesting ways: - Some let their agents migrate their New Relic dashboards to bitdrift. - Some automated their debugging end to end. For example, [ThredUp](https://bitdrift.io/case-study/thredup) now resolves 60% of mobile issues before a human ever sees them. - A large fintech built bitdrift into an internal Slack bot, so employees can ask questions that pull from bitdrift alongside their other internal tools. But there’s a common thread. All of these use cases run where engineers work: on a workstation, with the CLI installed, next to a coding agent. Meanwhile, more and more agents are running in the cloud, on their own, for hours at a time. Getting this set up in a consistent, reliable way turned out to be cumbersome with the existing tools. That’s why we built a remote MCP server. It opens bitdrift to agents running anywhere, and to everyone using them: product managers, support teams, marketers, and anyone else with a question about how the app is behaving or what users are experiencing. There’s nothing to install. You just point your agent to bitdrift’s MCP URL and your agents do the work, wherever they live, and answer you with real data. ## Your whole team can query reality Each of these used to mean filing a request with an engineer. Now it’s a question in a chat: - **Support:** “A customer says checkout froze on them this morning. What happened in their session?” - **Product:** “Is anyone using the new onboarding step, or are people skipping it?” - **Marketing:** “How many people opened the app from yesterday’s push notification, and where did they go next?” Every answer comes from what actually happened on your users’ devices. ## Your agents keep working while you’re away Because nothing runs on your machine, your agent doesn’t need you at your desk. Hand it a question before a meeting and read the answer after. Start a question on your phone and pick up the conversation on your laptop. It also means agents can work without you: watching for a crash spike after a release, digging in, and sending you what they found, down to the last screen users saw. ## Behind the scenes When you ask a connected agent a question, it works out what data it needs, queries bitdrift, and summarizes the result. All of that runs in the cloud. The only thing running on your phone is the chat. We chose MCP (Model Context Protocol) because it’s the open standard most AI apps already support, including ChatGPT, Claude, Codex, and Cursor. Your team can keep using their AI app of choice, and the MCP adds things a terminal can’t: you sign in with your bitdrift account, and interactive charts, sankeys, and 3D session replays show up right in the conversation, next to the evidence behind them. If your engineers already use the bitdrift CLI and skills, nothing changes. The CLI is still the most powerful way to churn through large volumes of data, and skills cover the entire platform. But the MCP server brings the power of bitdrift to diverse personas across their most common workflows, and we’ll continue expanding it. Use the CLI for depth, and MCP from anywhere. ## What your agent can see The agent has the same access as the person that connects it. If you can’t see an app’s data in bitdrift, your agent can’t either. ## Give it a try Ask your agent what’s happening with your app. Connecting just takes a couple of minutes: 1. Add `mcp.bitdrift.io` to your AI client or install the `bitdrift-plugin` from [our marketplace](https://github.com/bitdriftlabs/bd-skills). 2. Sign in to bitdrift when prompted. 3. Ask a question about your app. Setup steps for each client are in [the docs](https://docs.bitdrift.io/ai/mcp). New to bitdrift? [Start a free trial](https://bitdrift.io/signup). Query reality with bitdrift today. --- ## Frequently asked questions ### Do I need to install anything locally to ask an agent about my app? No. bitdrift’s MCP server is hosted in the cloud. You add its URL to an AI client you already use, such as ChatGPT or Claude, sign in and start asking questions. Nothing runs on your laptop or phone except the chat. ### Which AI tools work with bitdrift? Any AI client that supports remote MCP servers can connect to bitdrift’s mobile observability platform. That includes ChatGPT, Claude, Cursor, VS Code, Antigravity, Kiro, Copilot, Gemini, and OpenCode. [See the docs](https://docs.bitdrift.io/ai/mcp#apps-and-ides) for details. ### How is the bitdrift MCP server different from the bitdrift CLI and agent skills? All three give AI agents access to the same bitdrift data. The bd CLI and skills run on an engineer’s machine alongside a coding agent, and they’re the most powerful option for working through large volumes of data across the whole platform. bitdrift’s MCP server runs in the cloud, so anyone on a team can use it from a phone, a browser, or a cloud agent without installing anything.