claude-context-lint

> Audit your Claude Code setup and find exactly where tokens are being wasted.

[](https://claude.ai/code) [](https://www.npmjs.com/package/claude-context-lint) [](https://opensource.org/licenses/MIT)

Every Claude Code conversation starts with overhead: your CLAUDE.md files, skill descriptions, MCP tool schemas, and the base system prompt all consume context tokens before you type a single word. This tool makes that invisible cost visible.

npx claude-context-lint

Or install globally:

npm install -g claude-context-lint
claude-context-lint

  Claude Code Context Audit
  ────────────────────────────────────────────────────
  CLAUDE.md                1,240 tokens     █░░░░░░░░░░░░░░░░░░░
    CLAUDE.md                          1,240 tokens

  Skills (32 loaded)       4,800 tokens     ████░░░░░░░░░░░░░░░░  MEDIUM
    Listing overhead (per-turn):    4,800 tokens
    Full content (on invocation):   89,200 tokens
    ⚠ 3 near-duplicate skills detected (−420 tokens)
    api-helper                   120 listing      3,200 full
    db-migrate                   105 listing      2,800 full
    test-runner                   98 listing      1,950 full
    ... and 29 more

  MCP Servers (3)          14,100 tokens    █████████░░░░░░░░░░░  CRITICAL
    postgres               6,600 tokens   (22 tools) [always loaded]
    filesystem             1,800 tokens   (6 tools)  [always loaded]
    memory                 1,200 tokens   (4 tools)  [always loaded]

  System Prompt            8,500 tokens     █████░░░░░░░░░░░░░░░ (base overhead)
  ────────────────────────────────────────────────────
  TOTAL OVERHEAD:        28,640 tokens
  Context Limit:         200,000 tokens
  Used Befor