wiklob/linear-mcp-lean

linear-mcp-lean is a self hosted connector that lets AI agents like Claude Code talk to the project management tool Linear, using a standard called MCP that lets AI assistants call external tools. It sits in front of the same connector Linear itself offers, but it trims down what comes back from each request so an AI model reads far fewer words for the same task. The problem it solves is that Linear's own hosted connector returns full, bulky objects for every request, and every extra piece of returned data is something the AI model has to read and pay for in tokens, which adds up quickly across an agent session that might call these tools hundreds of times. This project rewrites each read so it only returns the fields an agent actually uses, shrinking a typical issue from around one kilobyte down to a few hundred bytes, and it makes write actions return a small receipt confirming what changed instead of echoing the whole updated item back. The tool names stay identical to Linear's own connector, so a workflow that already works can switch over without any other changes. If more detail is needed for a specific call, a user can ask for the full untrimmed response, or fall back to writing a raw query in the underlying data language directly, so trimming responses never becomes a dead end. The project also logs every call it makes so a person can check the actual measured savings on their own traffic through a built in statistics endpoint, rather than trusting a marketing claim. It authenticates with a simple personal access key from Linear rather than a more complex login flow, which makes it easy to run unattended in places like scheduled jobs or shared servers used by multiple automated agents. It can run locally as a single command with no server required, or be deployed once as a shared web service that multiple machines and agents can use at the same time. It requires a fairly recent version of Node.js and a personal Linear API key to get going, and is aimed at developers building or running AI coding agents that already talk to Linear and want to cut down on wasted cost.