Benchmark
Why context quality
beats context volume.
Seven real Odoo developer workflows. Each measured on four axes:
accuracy, full-codebase coverage, token savings, and speed.
Token counts are tiktoken cl100k_base,
MCP side live-captured on 2026-05-31.
Accuracy
Exact field/method names from the indexed graph — zero hallucination vs. agent guessing.
Full picture
Cross-repo blast radius — community + Viindoo + l10n modules aggregated, not just one repo.
Token savings
86–99% fewer tokens per workflow. Context window savings pay for the server in week one.
Speed
One MCP call replaces grep + multiple file reads. No manual scanning required.
7 real workflows · live-measured
Reproduce these numbers
All MCP-side token counts were captured live from the indexed server on 2026-05-31 using
tiktoken cl100k_base.
Without-MCP estimates follow the same tokenizer applied to grep output and file reads
that an AI agent would perform against a real ~/git/odoo17 checkout.
Full step-by-step accounting is in the raw data file.
To re-run: index your own Odoo addons with the OSM indexer, call the same MCP tools,
pipe the response text through tiktoken cl100k_base, and compare against
the grep + file-read baseline for your corpus.
Try it on your Odoo codebase
Connect your AI tool in under 5 minutes. No installation on your end.