MCP file tools silently eat your context window.I built one that doesnt
Hi, I am Anthony. Every token your filesystem tools consume is context the model cannot use for reasoning. Most MCP file servers are O(file size) on every operation: reads return the whole file, edits rewrite the whole file. The context window fills up before the agent gets anything meaningful done, and the problem compounds silently as your files grow. Chisel makes edits O(diff) and reads O(match). The agent sends a unified diff instead of a full rewrite, and queries with grep or sed instead of reading entire files. On a 500-line file this is nearly two orders of magnitude less context per…
What it does
In the maker’s words, at launch
Hi, I am Anthony. Every token your filesystem tools consume is context the model cannot use for reasoning. Most MCP file servers are O(file size) on every operation: reads return the whole file, edits rewrite the whole file. The context window fills up before the agent gets anything meaningful done, and the problem compounds silently as your files grow. Chisel makes edits O(diff) and reads O(match). The agent sends a unified diff instead of a full rewrite, and queries with grep or sed instead of reading entire files. On a 500-line file this is nearly two orders of magnitude less context per operation. The savings scale linearly with file size, so the larger your codebase, the more it matters. The other thing I cared about was security. Path confinement is enforced at the kernel level via cap-std and openat with O_NOFOLLOW, not a userspace prefix check. Shell commands run via direct execve against a fixed compile-time whitelist — no sh -c wrapper, no arbitrary execution. Atomic writes use a tmp-and-rename so a failed patch never corrupts the target file. It ships as a standalone MCP server, as chisel-core, a plain sync Rust library you can embed in your own MCP server, and as a WASM build for Node.js and Python runtimes, you can bring your own MCP and build on top of it. Would love feedback, especially from anyone building agents that do heavy file manipulation. GitHub: https://github.com/ckanthony/Chisel
Does the same job
all alternatives →
- MSMCP Server for Document Processing via Natural Language2025 · github.com · ▲8
Hey, I'm Nick from Nutrient, I want to share our newly released MCP Server that enables document workflows using natural language — things like redacting, merging, signing, converting formats, or extracting data. While many MCP servers have traditionally been developer-focused, we recognized that the technology could be highly effective in promoting the adoption of tools that are often hidden from end-user interfaces. We’re really interested to see if this side of the protocol could continue to mature. One thing we struggled with was the inability to receive documents from the client (no…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, March 2026
the whole month →
Switch from ChatGPT to Claude with import memory feature
AI · Mar 2026 · claude.com
