magi-linear-axi.
Context with intent.

I used Linear through MCP often. The integration worked, but the setup I was using loaded more than 20 tools and over 7,000 tokens of instructions before the agent had done any useful work.

That cost followed every conversation, whether the immediate task needed Linear or not. magi-linear-axi moves the capability behind one discoverable command-line interface. The agent can learn what it needs when it needs it, then return to the actual task.

The project was inspired by the thoughtful context-management ideas in the Agent eXperience Interface methodology.

The CLI covers common Linear work directly while keeping raw GraphQL available for the long tail.

• Modeled commands. Issues, teams, users, projects, cycles, milestones, initiatives, labels, documents, comments, relations, and updates have explicit command paths.
• Compact output. TOON is the default, with compact JSON available for downstream consumers.
• Predictable automation. No interactive prompts, stable exit codes, bounded reads, and errors designed for recovery.
• Deliberate safety. Read-only operations may retry transient failures; mutations and ambiguous operations do not retry blindly.
• Escape hatch included. Raw GraphQL remains available when a modeled command is not enough.

A published v0.2.1 benchmark compared compact, read-only agent workflows against the official Linear MCP.

Read the benchmark ↗

Install the CLI, provide a Linear personal API key through the environment, and verify the authenticated user.

cargo install magi-linear-axi --locked
export LINEAR_API_KEY='lin_api_your_key'
magi-linear-axi auth whoami

Open the documentation ↗