MCP Token Ledger
Every MCP server you connect injects its tools/list response into the model context on every turn. This page measures that cost so you can budget your window before the model starts drifting. Heavy servers are often the reason a stack feels sluggish — a single 7,000-token MCP eats more context than most conversations.
| Server | Tools | Tokens | Cost |
|---|---|---|---|
| GitHub (Official) DEV_TOOLS | 27 | 7,400 | Heavy |
| Playwright (Microsoft) BROWSER | 32 | 6,800 | Heavy |
| Stripe INTEGRATION | 24 | 5,100 | Medium |
| Jira (Atlassian) DEV_TOOLS | 24 | 4,800 | Medium |
| Linear PRODUCTIVITY | 22 | 4,200 | Medium |
| Slack COMMUNICATIONS | 18 | 3,800 | Medium |
| Filesystem FILESYSTEM | 14 | 3,493 | Medium |
| Notion PRODUCTIVITY | 16 | 3,400 | Medium |
| Google Drive FILESYSTEM | 14 | 3,100 | Medium |
| Memory MEMORY | 9 | 2,802 | Light |
| Gmail COMMUNICATIONS | 11 | 2,700 | Light |
| Sentry DEV_TOOLS | 12 | 2,400 | Light |
| Supabase DATABASE | 10 | 2,200 | Light |
| sequentialthinking | 1 | 1,287 | Light |
| Postgres DATABASE | 6 | 1,100 | Light |
| Puppeteer BROWSER | 7 | 699 | Trivial |
| Brave Search SEARCH | 2 | 415 | Trivial |
| everart | 1 | 255 | Trivial |
Harness measurements spawn each MCP server over stdio, send the initialize and tools/list JSON-RPC messages, and count tokens in the serialised tools response. Tokens are estimated at roughly 3.5 characters per token — consistent across servers, within 15% of exact tokenizer output, and good enough for ranking.
Editorial estimates are used for servers we can't spawn locally (commercial SaaS, auth-required, Python-only). They get replaced the moment we have real harness data.
Why this matters: context is a zero-sum resource. A 6,000-token MCP on a 200K-context model costs 3% of your window every turn, every conversation, forever. Stacking three heavy MCPs can burn 10%+ of context before the model has read a single user message.