Benchmark — May 2026
Token cost: inline vs. Botverse MCP
Processing a 2MB research PDF inline vs. offloading to Botverse via MCP. Measured with claude-sonnet-4-6.
Inline (text extraction)
97,354
input tokens
- ✕Extract PDF text via pdftotext
- ✕Pass raw text to Claude context
- ✕Agent reads every word, header, footnote
- ✕Context window dominated by file content
12× fewer tokens
Botverse MCP
~8,000
input tokens
- ✓Agent calls convert_from_url MCP tool
- ✓Conversion runs server-side in ~3 seconds
- ✓Agent receives a download URL, not content
- ✓File size has no effect on token count
Relative token usage
Botverse MCP~8,000 tokens
At scale — 500 documents/month
Methodology: A 2MB research PDF was processed in two modes using claude-sonnet-4-6. Inline mode: text extracted via pdftotext and passed directly to the model context. Botverse mode: agent calls convert_from_url, get_job_status, and get_download_url — total MCP tool exchange only. Token counts measured via the Anthropic API usage object. Pricing at $3.00/MTok input. Botverse convert fee: $0.05/job.
Every token your agent spends on file I/O is wasted reasoning capacity.