Developers distracted 1,200 times daily, MCP targets productivity fix

Software developers spend far more time not coding than they do writing actual code. Industry studies suggest direct programming may account for only around 16% of a developer's workday, with the vast majority of time eaten up by supporting tasks and operations. In an era where teams are pushed to achieve more with fewer resources and CEOs tout their AI-generated code, a critical question emerges: how are we optimizing the other 84% of an engineer’s workload?

Keep Developers Where They Are Most Productive

One of the biggest drains on developer productivity is context switching—the constant jumping between the ever-expanding suite of tools and platforms required to build and ship software. According to a Harvard Business Review study, the average digital worker switches between apps and websites nearly 1,200 times daily. Every interruption carries a cost. Research from the University of California indicates it takes roughly 23 minutes to fully regain focus after a single disruption, and worse, nearly 30% of interrupted tasks are never finished. This challenge of context switching is so central that it's woven into DORA, one of the most widely adopted frameworks for measuring software development performance.

As AI-driven organizations seek to empower employees to do more with less—going beyond simply providing access to large language models (LLMs)—new trends are taking shape. Jarrod Ruhland, a principal engineer at Brex, believes that "developers deliver their highest value when they stay focused within their integrated development environment (IDE)." Guided by this principle, he explored new methods to make this a reality, and Anthropic's new protocol might hold a key piece of the puzzle.

MCP: A Protocol to Bring Context to IDEs

AI-powered coding assistants, such as LLM-enhanced IDEs like Cursor, Copilot, and Windsurf, are fueling a developer renaissance, and their adoption rate is unprecedented. Cursor became the fastest-growing SaaS product in history, reaching $100 million in annual recurring revenue within a year of launch, while 70% of Fortune 500 companies now use Microsoft Copilot.

However, these assistants have traditionally been confined to the codebase context. While this helps developers write code faster, it doesn't address the broader problem of context switching. A new protocol aims to solve this: the Model Context Protocol (MCP). Launched by Anthropic in November 2024, MCP is an open standard designed to simplify integration between AI systems—especially LLM-based tools—and external data sources and applications. Its popularity is surging, with a 500% increase in new MCP servers over the past six months and an estimated 7 million downloads in June alone.

One of MCP's most impactful applications is its ability to connect AI coding assistants directly to the everyday tools developers use, streamlining workflows and dramatically cutting down on context switching.

Consider feature development. Traditionally, this involves hopping between multiple systems: reading the ticket in a project tracker, reviewing a teammate's conversation for clarity, searching documentation for API specifics, and finally opening the IDE to start coding. Each step exists in a separate tab, forcing mental shifts that slow down progress.

With MCP and modern AI assistants like Anthropic's Claude, this entire process can unfold inside the code editor.

For example, implementing a feature entirely within a coding assistant could look like this:

• Pull in ticket details using a Linear MCP server;
• Surface relevant discussions using a Slack MCP server;
• Access the necessary documentation via a Glean MCP server;
• Write the feature by asking Cursor to generate the initial scaffolding.

The same approach applies to other engineering workflows. An incident response for site reliability engineers (SREs), for instance, might involve:

• Pulling incident details via a Rootly MCP server
• Retrieving trace data through a Sentry MCP server
• Importing observability metrics via a Chronosphere MCP server
• Resolving the bug by instructing Claude Desktop

Nothing New Under the Sun

We've seen this pattern before. Over the last decade, Slack revolutionized workplace productivity by becoming a central hub for hundreds of apps, allowing employees to manage diverse tasks without ever leaving the chat window. Slack's platform successfully reduced context switching in daily workflows.

Take Riot Games as an example. By integrating around 1,000 Slack apps, their engineers reported a 27% reduction in the time needed to test and iterate on code, identified new bugs 22% faster, and increased their feature launch rate by 24%. These gains were largely attributed to streamlined workflows and less friction from switching between tools.

Now, a similar transformation is unfolding in software development. AI assistants, empowered by MCP integrations, are becoming the bridge to all these external tools. In effect, the IDE is poised to become the new all-in-one command center for engineers, much as Slack became for general knowledge workers.

MCP May Not Be Enterprise-Ready

MCP is a relatively young standard. From a security perspective, it lacks built-in authentication or a permission model, relying instead on external implementations that are still evolving. There's also ambiguity around identity and auditing—the protocol doesn't clearly distinguish whether an action was initiated by a user or the AI itself, complicating accountability and access control without custom solutions. Lori MacVittie, a distinguished engineer and chief evangelist in F5 Networks' Office of the CTO, notes that MCP is "breaking core security assumptions that we've held for a long time."

Another practical limit emerges when too many MCP tools or servers are used simultaneously within a coding assistant. Each MCP server advertises a list of available tools, complete with descriptions and parameters, that the AI model must process. Flooding the model with dozens of tools can overwhelm its context window, causing noticeable performance degradation as the tool count grows. Some IDE integrations have implemented hard limits—around 40 tools in Cursor IDE, or roughly 20 for the OpenAI agent—to prevent prompts from ballooning beyond the model's capacity.

Finally, there is currently no sophisticated method for tools to be auto-discovered or suggested contextually; they are simply listed. This often forces developers to manually enable or disable tools or curate active sets to keep workflows smooth. Considering the earlier example of Riot Games installing 1,000 Slack apps, it's clear how an unmanaged proliferation of tools could be problematic for enterprise use.

Less Swivel-Chair, More Software

The past decade has demonstrated the value of bringing work to the worker—from Slack channels that centralize updates, to "inbox zero" email strategies, and unified platform engineering dashboards. Now, with AI added to our toolkit, we have a prime opportunity to supercharge developer productivity. If Slack became the hub for business communication, then AI coding assistants are perfectly positioned to become the hub for software creation—not just where code is written, but where all relevant context and collaboration converge.

By allowing developers to remain in their productive flow, we eliminate the constant mental gear-shifting that has long hampered engineering output.

For any organization that relies on software delivery, it's worth taking a hard look at how your developers actually spend their time. You might be surprised by what you discover.

Sylvain Kalache leads AI Labs at Rootly.