Skip to content

docs: fix drift in ChatGPT - #266

Draft
jack-arturo wants to merge 1 commit into
mainfrom
docs/audit-platforms-chatgpt-20260727
Draft

docs: fix drift in ChatGPT#266
jack-arturo wants to merge 1 commit into
mainfrom
docs/audit-platforms-chatgpt-20260727

Conversation

@jack-arturo

Copy link
Copy Markdown
Member

Routine docs-accuracy audit of src/content/docs/docs/platforms/chatgpt.md (never previously reviewed — no entry in scripts/file-doc-map.json).

One finding, stated in two places. It is the last unswept instance of the "Flask defaults to 5000" myth: the same false claim was removed from overview, operations/troubleshooting, and deployment/railway (#247), and from platforms/claude-web in the companion PR from this run (#265).

Claim Current state Fix Evidence
Railway :::caution — "The memory service must have PORT=8001 set — without it, Flask defaults to port 5000 and connections will fail." The service defaults to 8001: port = int(os.environ.get("PORT", "8001")). There is no code path that produces 5000. Leaving PORT unset is the working configuration, so this instructs users to set a variable to avoid a failure that cannot happen. Rewrote to the real constraint: the service listens on 8001 unless PORT is overridden, and the port in AUTOMEM_API_URL has to match whatever it is actually bound to. runtime_wiring.py#L82
"Memory service connection errors (502/504)" cause table — first row: "Missing PORT=8001 in memory service" → "Add PORT=8001 to memory service environment variables" Same false premise, and here it actively misdiagnoses: a reader hitting ECONNREFUSED on :8001 will add a variable whose value already applies, conclude the fix didn't work, and still not know the real cause. Replaced the row with "Memory service bound to a different port", pointing at the actual check — compare PORT on the service against the port in AUTOMEM_API_URL. same as above

Spot-checked and correct, left untouched:

  • All four sidecar endpoints exist in mcp-sse-server/server.js: /mcp (L1039), /mcp/sse (L1092), /mcp/messages (L1126), /health (L857).
  • Sidecar PORT default 8080 (const port = process.env.PORT || 8080), and both AUTOMEM_API_URL and AUTOMEM_API_TOKEN are read as documented.
  • 20-second SSE heartbeat (setInterval(…, 20000)).
  • All six tool names match the tools array in mcp-automem's src/index.ts.
  • The unauthenticated curl …/health in Troubleshooting — the sidecar's /health route takes no request token; it probes upstream with its own AUTOMEM_API_TOKEN.
  • The ${{...}} Railway variable-resolution section and the "Security Notes" bullets about URL tokens appearing in logs.

Verified against: verygoodplugins/automem@8ff266e

Questions

None.

Unverified

  • "ChatGPT requires URL-based token authentication for custom MCP connectors. Header-based auth is not supported by ChatGPT's connector configuration UI." — an OpenAI product constraint with no in-repo counterpart. It is load-bearing for this page's whole approach, so it is worth re-confirming against ChatGPT's current Developer Mode UI, but I can't check it from here and did not touch it.
  • Connector setup navigation (Settings → Connectors → Advanced → Developer Mode → + Add Server) and the "Plus, Pro, Team, or Enterprise" tier requirement.
  • Railway one-click template contents and the Generate Domain flow.

Follow-ups

Below the severity bar for this PR:

  • The Sidecar endpoints table lists /mcp as method POST; it is registered with app.all('/mcp', …), so GET and DELETE also reach it. POST is the right thing to document — the table is just narrower than the route. Same note applies to the identical table on platforms/claude-web.

Generated by Claude Code as part of the recurring automem-docs-review routine.


Generated by Claude Code

Remove the false "Flask defaults to port 5000" claim from the Railway
caution and rewrite the ECONNREFUSED troubleshooting row that was
premised on it. The memory service defaults to port 8001.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Bh4cvNW2mR2vCoyg3FDQL3
@cloudflare-workers-and-pages

Copy link
Copy Markdown

Deploying automem-website with  Cloudflare Pages  Cloudflare Pages

Latest commit: a25c8c4
Status: ✅  Deploy successful!
Preview URL: https://3ddb2ae5.automem-website.pages.dev
Branch Preview URL: https://docs-audit-platforms-chatgpt.automem-website.pages.dev

View logs

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants