TODO: Fix pal-e-docs-mcp PyPI publish pipeline failure

todo-fix-mcp-pypi-publish Todo

todo done

Project

Problem

The publish step in forgejo_admin/pal-e-docs-mcp Woodpecker pipeline fails on every push to main. Clone and lint pass. Publish exits with code 1.

What We Know

Clone fix (PR #15) works — internal Forgejo URL resolves from pipeline pods
Lint step passes
PyPI URL was changed from from_secret to hardcoded internal URL in PR #17
Internal Forgejo PyPI endpoint responds correctly (405 on GET, allows POST) — verified via curl from cluster
Package version is 0.1.0 — never changes between builds
Woodpecker log streaming is broken (empty newlines) — cannot read actual error output

What We Don't Know

The actual error message from twine — logs are empty due to Woodpecker 3.13.0 log streaming bug
Could be: DNS resolution failure in the publish container (different from clone container)
Could be: version conflict (0.1.0 already published, twine rejects duplicate)
Could be: auth failure (secrets not injected properly after URL change)
Could be: something else entirely

Impact

Low — pal-e-docs-mcp runs locally from source via uv run. The published PyPI package is not consumed by anything in production. But the pipeline should work.

To Investigate

Fix Woodpecker log streaming so we can actually read errors (may require Woodpecker DB cleanup or version upgrade)
Or: add set -x / verbose output to the publish commands to force output to stdout before the log stream
Or: run a test pod in the woodpecker namespace that exactly replicates the publish step (python:3.12-slim, pip install build twine, build, upload)
Check if version 0.1.0 is already in the Forgejo PyPI registry
Consider adding auto-version-bump (e.g. using git SHA or date) to avoid conflicts