VercelPyCheck — find the Vercel Python deploy bug before it 500s in prod

Vercel's Python runtime will happily ship a green build that 500s in production with no traceback. Paste your project below for a free instant check, or scan a full public GitHub repo automatically for $9.

Free instant check

Paste your requirements.txt/pyproject.toml and your api/*.py files below, one per ### FILE: path marker line, for example:

### FILE: requirements.txt
flask

### FILE: api/index.py
from flask import Flask
app = Flask(__name__)

What gets checked

requirements.txt + pyproject.toml conflict — Vercel's Python builder now prefers pyproject.toml (uv) over requirements.txt with no lockfile present; a confirmed live regression (vercel/vercel#14041) that silently skips installing your runtime dependencies.
Manifest at the wrong path — a requirements.txt sitting only inside api/ instead of the project root is never read.
Sibling-module imports without sys.path — Vercel's Python runtime doesn't add a function's own directory to sys.path, so importing a sibling file the normal way 404s or 500s with zero traceback shown anywhere.
Multiple top-level ASGI apps under api/ — confirmed to route reliably in some projects and silently swallow every route but one in others, undocumented and build-config-dependent.
Streaming/SSE ASGI apps — some streaming response patterns have been observed to crash with FUNCTION_INVOCATION_FAILED on Vercel's Python runtime even on a plain GET.
Unzipped size limit risk — flags dependencies (torch, tensorflow, scipy, opencv-python, transformers) that commonly push a function over Vercel's 250MB unzipped ceiling.
.well-known/ files with no route — a domain-verification or MCP-registry proof file placed under .well-known/ with no vercel.json or explicit route can be silently unreachable even though it's genuinely in the deployment. (Ordinary root files like robots.txt are unaffected — this is specific to dotfile directories.)

Nothing you paste is stored. It's used in memory for the single request that generates your report, then discarded. The $9 repo scan only reads a public GitHub repo's api/*.py files and root dependency manifests over GitHub's public API — no token, no clone, no access to private repos.

Privacy · Terms

Other tools from the same builder: Edge Thirteen (value-investing newsletter) · StripeCheckup (Stripe billing-logic audit) · TriageShield (AI-slop vuln-report checker) · FilingCheck (SEC-grounded stock reports) · BrewClear (TTB label clearance) · MarkRadar (trademark + domain-squat alerts) · ExploitWatch (EU CRA exploit alerts) · AuditFlag (SEC going-concern alerts) · RenewalMatrix (click-to-cancel compliance) · SlopCheck (YouTube demonetization-risk checker)