Summary
Two README ↔ code drifts found during cold-turkey testing:
1. update-secrets.sh and hf-secret
README §"Installation" (line 191) says update-secrets.sh "Creates/updates hf-secret (skips token if not set)". The actual script skips entirely if HF_TOKEN is unset, leaving hf-secret absent. On kind this causes:
CreateContainerConfigError: secret "hf-secret" not found
Either the README is wrong (it should say the secret is only created when HF_TOKEN is set) or the script should create a placeholder secret so the MCP pod can start.
2. OPENAI_API_BASE documented as optional but practically mandatory
OPENAI_API_BASE is documented on line 79 as the variable to point at your endpoint, but listed on line 201 as "optional". When using an OpenAI-compatible proxy (which the default openai/Azure/gpt-4.1 model requires), leaving it unset causes every LLM call to return Incorrect API key even after update-secrets.sh has patched the key correctly.
The README should clarify: "OPENAI_API_BASE is required when using an OpenAI-compatible proxy."
Found in
PR #33 review by @kellyaa.
Summary
Two README ↔ code drifts found during cold-turkey testing:
1.
update-secrets.shandhf-secretREADME §"Installation" (line 191) says
update-secrets.sh"Creates/updateshf-secret(skips token if not set)". The actual script skips entirely ifHF_TOKENis unset, leavinghf-secretabsent. On kind this causes:Either the README is wrong (it should say the secret is only created when
HF_TOKENis set) or the script should create a placeholder secret so the MCP pod can start.2.
OPENAI_API_BASEdocumented as optional but practically mandatoryOPENAI_API_BASEis documented on line 79 as the variable to point at your endpoint, but listed on line 201 as "optional". When using an OpenAI-compatible proxy (which the defaultopenai/Azure/gpt-4.1model requires), leaving it unset causes every LLM call to returnIncorrect API keyeven afterupdate-secrets.shhas patched the key correctly.The README should clarify: "
OPENAI_API_BASEis required when using an OpenAI-compatible proxy."Found in
PR #33 review by @kellyaa.