Cloud-v1 tier parity harness
July 18, 2026 · View on GitHub
Compares @erpipe/core pure tools against mcp-odoo (Python) on a shared fixture corpus. This is cloud-v1 tier parity (D14: 23 tools / 7 prompts), not full local-server parity (41 tools).
Run
# requires sibling mcp-odoo checkout with importable odoo_mcp
MCP_ODOO_PATH=../mcp-odoo PYTHON=python3 npm run parity
# or use the project venv
MCP_ODOO_PATH=../mcp-odoo PYTHON=../mcp-odoo/.venv/bin/python npm run parity
# TS surface + render only (no Python)
npm run parity:ts
What is compared
| Tool / surface | Mode |
|---|---|
| Surface 23 tools + 7 prompts | exact name set |
generate_json2_payload | deep JSON |
upgrade_risk_report | deep JSON |
fit_gap_report | deep JSON |
business_pack_report | deep JSON |
build_domain | deep JSON |
preview_write | approval token + issues codes (choke point) |
| 3 sample prompts | exact text |
Live Odoo tools (search/read/profile/schema/write execute) are covered by npm run smoke:live, not this pure harness.
Files
scripts/parity/corpus.json— cases + surface manifestscripts/parity/run_python_side.py— Python fixture runnerscripts/parity-harness.mjs— orchestrator + normalize/diff
Latest
2026-07-18: 16/16 pure cases PASS vs mcp-odoo on this machine.