New Tool Cuts Claude Code Context Bloat by 98%, Extending AI Session Time Threefold

A groundbreaking open-source tool named Context Mode is transforming how developers interact with AI coding assistants, particularly Anthropic’s Claude Code. Developed by engineer mksglu and released under the MIT license, Context Mode acts as an intelligent intermediary between Claude Code and its external tooling interfaces—commonly known as Model Control Protocol (MCP) calls—significantly reducing the volume of raw data consumed by the AI’s context window. According to the project’s GitHub repository, a typical 315 KB payload from multiple Playwright snapshots and GitHub issue dumps is compressed into a mere 5.4 KB of distilled summary data, representing a 98.3% reduction in context usage.

The innovation addresses a critical bottleneck in AI-assisted development: context window saturation. As developers integrate tools like Playwright for browser automation, GitHub API fetches, and local file system scans, each output floods Claude Code’s 200 KB context limit. After just 30 minutes of continuous use, up to 40% of the context window becomes occupied by redundant or low-value data, forcing the AI to forget earlier instructions and degrade performance. Context Mode solves this by intercepting these outputs in a sandboxed environment, processing them with SQLite FTS5 and BM25 ranking algorithms to extract only the most relevant information, then returning concise, semantically rich summaries.

The tool supports ten major programming language runtimes—including Python, Node.js, Go, and Rust—and enables batch execution of multiple MCP calls in a single pass. This not only reduces latency but also ensures consistency across multi-step workflows. Developers report that session longevity has increased from approximately 30 minutes to over three hours without performance degradation, enabling more complex, uninterrupted coding sessions.

Installation is streamlined via a single command: /plugin marketplace add mksglu/claude-context-mode followed by /plugin install context-mode@claude-context-mode. The project’s documentation includes detailed benchmarks comparing raw vs. summarized outputs across real-world scenarios, such as parsing 20 GitHub issues (reduced from 59 KB to 1.1 KB) and capturing Playwright snapshots (down from 56 KB to 0.8 KB). These figures underscore the tool’s efficiency in preserving semantic meaning while eliminating noise.

On Hacker News, where the project was first shared, the announcement garnered 50 points and 15 comments within two hours, with developers praising its simplicity and impact. One user noted, "This is the missing piece for anyone doing heavy automation in Claude Code. I’ve been manually trimming outputs—this automates it perfectly." Another highlighted its potential for enterprise environments where long-running AI sessions are critical for debugging complex systems.

Unlike other context optimization techniques that rely on model fine-tuning or external APIs, Context Mode operates locally and requires no subscription or cloud dependency. Its open-source nature invites community contributions, and the MIT license ensures broad adoption across commercial and academic projects. The project’s GitHub page includes a roadmap for future enhancements, including support for LLM-driven summarization of non-textual data like logs and JSON schemas.

As AI coding assistants become central to modern software development workflows, tools like Context Mode represent the next evolution: not just smarter AI, but smarter data management. By decoupling raw output from context consumption, mksglu’s innovation empowers developers to work with greater speed, depth, and endurance—turning what was once a technical constraint into a scalable advantage.