Premium requests drain fast
Agent-mode sessions consume premium requests while the model wades through raw file dumps and full terminal logs.
LeanCTX for GitHub Copilot
Copilot's agent mode, without the context tax.
LeanCTX cuts GitHub Copilot agent-mode token usage by 60–90% through MCP: compressed AST-aware reads, ~13-token cached re-reads and 88–99% smaller terminal output. Premium requests go further because each one carries signal instead of raw dumps. Setup: lean-ctx setup, auto-detected.
The problem
Where GitHub Copilot spends your tokens.
Workspace context is shallow
Copilot sees fragments. Without a structural layer, it rereads files to rebuild understanding it already had.
Every IDE is configured separately
VS Code, JetBrains, CLI: each integration is its own setup. A context layer should configure once and serve all.
Your stack. Your agent. Better context.
The integration
How LeanCTX plugs into GitHub Copilot.
MCP server for Copilot
lean-ctx setup registers the MCP server where Copilot looks for it (VS Code and compatible IDEs). Agent mode picks up ctx_read, ctx_search and ctx_shell.
Copilot instructions installed
Setup writes the instructions block so agent mode prefers compressed tools without manual prompting.
One layer for every editor
The same local binary serves VS Code, JetBrains and 28 other tools. One cache, one ledger, one configuration.
Setup
One command. Auto-detected.
terminal
# install LeanCTX
$ curl -fsSL https://leanctx.com/install.sh | sh# auto-detect Copilot-capable IDEs and configure MCP
$ lean-ctx setup# verify the wiring
$ lean-ctx doctor# see what agent mode saved
$ lean-ctx gainWhat changes
Three moments you'll notice.
Premium requests that go further
Each request carries 60–90% less dead weight, so multi-step agent tasks finish within budget instead of timing out on context.
The repo map instead of the repo
ctx_read in map mode hands Copilot the structure (exports, signatures, dependencies) for a fraction of a full read.
One number for the team
The signed ledger aggregates savings across every developer's machine, exportable when finance asks what changed.
GitHub Copilot + LeanCTX, answered.
Install LeanCTX and run lean-ctx setup. Copilot's agent mode gets MCP tools that compress reads by 60–90% and cache re-reads at ~13 tokens, each premium request does more work.
VS Code and JetBrains IDEs with MCP-capable Copilot, plus 28 other AI tools, all served by the same local binary after one setup command.
No. LeanCTX is local-first with zero telemetry. It compresses locally and hands the result to Copilot. No additional egress, ever.
Using this at work?
Team gives your whole org one shared, audited context plane, and the ledger proves the savings to finance. Local use stays free forever, enforced by CI.