{"slug": "claude-code-is-burning-your-token-budget-here-s-the-receipt", "title": "Claude Code Is Burning Your Token Budget. Here's the Receipt.", "summary": "A developer discovered that MCP servers were silently consuming up to 111,000 tokens per conversation in Claude Code, wasting an estimated $2,500 per year. By auditing and removing unused servers, they reduced context overhead by 97% and tripled conversation length. The developer released a tool called mcptoon to compress MCP schemas.", "body_md": "I found $2,500/year of hidden token waste in my Claude Code setup. It was the MCP servers.\n\nLast week I noticed my Claude Code conversations were dying at around message 15. Context window full. The model starts forgetting earlier instructions. Tool calls fail. The conversation degrades into hallucination.\n\nI assumed it was my fault — too many messages, too much context. So I started measuring.\n\nHere's what I found:\n\n```\nSession start:\n  Claude system prompt:        ~8,000 tokens\n  MCP schema injection:      ~111,000 tokens\n  User's first message:            50 tokens\n  ──────────────────────────────────────────\n  Total before any work:     ~119,000 tokens\n\n  Remaining context:         ~81,000 tokens\n```\n\n**I was starting every conversation with 60% of my context already consumed.**\n\nThe culprit wasn't my prompts. It was the 10 MCP servers I had proudly configured in my `claude_desktop_config.json`\n\n.\n\nI measured each server's schema injection using tiktoken:\n\n| Server | Why I Installed It | Token Cost | Times Used/Week |\n|---|---|---|---|\n| GitHub | PR reviews, issues | 12,440 | 3 |\n| Slack | Message reading | 14,672 | 0 |\n| Google Drive | Doc access | 47,293 | 1 |\n| Notion | Knowledge base | 13,780 | 2 |\n| Postgres | Query DB | 8,231 | 4 |\n| Puppeteer | Screenshots | 5,890 | 0 |\n| Filesystem | File access | 3,847 | 15 |\n| Brave Search | Web search | 2,103 | 5 |\n| Memory | Context persistence | 2,567 | 0 |\n| Sequential Thinking | Reasoning | 890 | 2 |\nTotal |\n111,713 |\n\nLook at the \"Times Used/Week\" column. **Three servers were used zero times.** Two more were used once or twice. But every single one of them was injecting 100% of its schema into every conversation.\n\nI was paying $0.33 per conversation — $2,500/year — to load schemas for tools I barely used.\n\nI posted my findings on Bluesky. Within hours:\n\n\"I had the same issue. Removed 6 MCP servers and my conversations went from dying at message 15 to lasting 40+ messages.\" —\n\n[@developer1]\"GitHub MCP is 12K tokens but Claude Code already has\n\n`gh`\n\nCLI built in. Why did I install it?\" — @developer2\"Google Drive MCP alone is 47K tokens. FORTY SEVEN THOUSAND. For a tool I used once this month.\" — @developer3\n\nThis isn't a niche problem. **Everyone running 5+ MCP servers is silently burning 50-100K tokens per conversation.**\n\n```\n# Count token cost of each server\npip install mcptoon\nmcptoon audit --config ~/.config/claude/claude_desktop_config.json\n```\n\nOutput:\n\n```\nFilesystem:     3,847 tokens  ✓ Keep (used daily)\nGitHub:        12,440 tokens  ✗ Remove (use `gh` CLI instead)\nGoogle Drive:  47,293 tokens  ✗ Remove (use `gdrive` CLI)\nSlack:         14,672 tokens  ✗ Remove (use `slack` CLI)\nNotion:        13,780 tokens  ⚠ Depends (no good CLI alternative)\nPostgres:       8,231 tokens  ✓ Keep (used 4x/week)\n...\nTotal waste:   89,702 tokens  →  $1,346/year\n{\n  \"mcpServers\": {\n    \"filesystem\": { ... },\n    \"postgres\": { ... },\n    \"brave-search\": { ... }\n  }\n}\n```\n\nI went from 10 servers to 3. My context overhead dropped from 111K to 14K tokens.\n\nFor servers you must keep, wrap them with mcptoon to compress schemas:\n\n```\n{\n  \"mcpServers\": {\n    \"postgres\": {\n      \"command\": \"mcptoon\",\n      \"args\": [\"serve\", \"--stdio\", \"npx\", \"@modelcontextprotocol/server-postgres\"]\n    }\n  }\n}\n```\n\nmcptoon compresses tool schemas by 97% using a compact TOON format. The model sees the same tools but with 3K tokens instead of 111K.\n\n| Metric | Before | After | Change |\n|---|---|---|---|\n| Context at session start | 111K tokens | 3.2K tokens | -97% |\n| Conversations before context death | ~15 messages | ~45 messages | 3x |\n| Token cost per conversation | $0.54 | $0.14 | -74% |\n| Monthly cost (20 conv/day × 22 days) | $237 | $62 | -$175 |\n| Annual savings | — | — | $2,100 |\n\nThis isn't just about money. The real cost of MCP schema bloat is **cognitive degradation**:\n\nWhen Perplexity's CTO announced they were replacing MCP with REST API + CLI internally, this is what he was talking about. When Garry Tan said \"MCP sucks honestly,\" this is what he meant. When Anthropic's own engineers showed 98.7% token reduction by not loading schemas into context, this is the problem they were solving.\n\nMCP servers aren't bad. The protocol isn't bad. But **loading every tool schema into context at startup is a design flaw that costs every developer thousands of dollars per year in wasted tokens.**\n\nMost developers don't know this because:\n\nNow you know. Go audit your config.\n\n```\npip install mcptoon\nmcptoon audit --config your-config.json\n```\n\n*mcptoon is open source, Apache 2.0, 250KB, zero dependencies. GitHub · PyPI. All measurements use tiktoken cl100k_base. Not affiliated with Anthropic.*", "url": "https://wpnews.pro/news/claude-code-is-burning-your-token-budget-here-s-the-receipt", "canonical_source": "https://dev.to/mcptokensaver/claude-code-is-burning-your-token-budget-heres-the-receipt-1nkf", "published_at": "2026-08-23 09:18:35+00:00", "updated_at": "2026-08-23 09:43:12.985542+00:00", "lang": "en", "topics": ["developer-tools", "artificial-intelligence", "large-language-models", "ai-infrastructure"], "entities": ["Claude Code", "MCP", "mcptoon", "GitHub", "Google Drive", "Slack", "Notion", "Postgres"], "alternates": {"html": "https://wpnews.pro/news/claude-code-is-burning-your-token-budget-here-s-the-receipt", "markdown": "https://wpnews.pro/news/claude-code-is-burning-your-token-budget-here-s-the-receipt.md", "text": "https://wpnews.pro/news/claude-code-is-burning-your-token-budget-here-s-the-receipt.txt", "jsonld": "https://wpnews.pro/news/claude-code-is-burning-your-token-budget-here-s-the-receipt.jsonld"}}