Skip to content

docs: verify real clients and expand task guides - #1741

Merged
Dumbris merged 1 commit into
mainfrom
codex/docs-real-client-guides
Oct 22, 2026
Merged

Dumbris merged 1 commit into
mainfrom
codex/docs-real-client-guides

Conversation

@Dumbris

@Dumbris Dumbris commented Oct 22, 2026 •

Copy link
Copy Markdown
Member

Problem and change

The first task guides had only protocol-level validation, the introduction listed assumed client brands, and there were no complete workflows for comparing discovery with direct routing or reviewing a changed tool.

Validate the existing setup and diagnosis guides in real Claude Code and Codex CLI sessions, correct their recovery guidance, and add two tested guides: Keep a growing tool catalog out of initial context and Review a tool before exposing it to your agent. Both new guides state a measurable outcome, prerequisites, exact commands, expected results, recovery and a tested-version stamp. Wire all four guides into the landing page and sidebar.

Replace the introduction's client list with names from a live, read-only, non-CI telemetry snapshot for September 22–October 22, 2026. The text distinguishes configured integrations from self-reported client names (including earlier connections), and separates Codex CLI from ChatGPT and VS Code from Copilot. Correct the introduction's stale quarantine exceptions against released v0.71.1 source.

Validation

  • Real MCPProxy v0.71.1 in scratch storage on high ports; production daemon and normal AI-client MCP configurations were not changed.
  • Claude Code 2.1.307 and Codex CLI 0.159.2: actual retrieve_tools → call_tool_read calls to Context7 returned library matches; Activity attributes the calls to distinct client credentials.
  • Both clients' unavailable-tool diagnosis: deliberate scratch tool disable, preflight tool_blocked_by_user, real TOOL_BLOCKED error, then successful retry after re-enable.
  • Discovery guide: exact copyable probe run against a controlled catalog; both discovery and direct routes completed an echo call. Byte counts are explicitly transport measurements, not token/cost claims.
  • Review guide: selected one of 24 tools for first approval; a later description change was held as tool_changed and remained held after a fresh clean scan. Documents that blocked sibling definitions can remain listed in direct mode in v0.71.1 while calls are refused.
  • go test ./cmd/release-gate ./internal/connect -count=1.
  • npm --prefix website run build, including artifact verification of all 134 published LLM links.
  • Independent fresh-context review and deployed preview verification recorded below after completion.

Independent fresh-context review: VERDICT: PASS, no remaining actionable findings. Final production docs build and release-gate/connect rerun passed after all wording and validation stamps were finalized. Browser inspection confirmed the new procedures and four-guide navigation render correctly.

Deployed preview

Expanded guides preview verified live: all 134 exported routes return successful responses; the four guides and landing page have descriptions and canonical URLs and appear in the sitemap and LLM exports. The telemetry-derived client list replaces the old combined ChatGPT/Codex label.

@cloudflare-workers-and-pages

Copy link
Copy Markdown

Deploying mcpproxy-docs with  Cloudflare Pages  Cloudflare Pages

Latest commit: 3b4ee04
Status: ✅  Deploy successful!
Preview URL: https://df2793c4.mcpproxy-docs.pages.dev
Branch Preview URL: https://codex-docs-real-client-guide.mcpproxy-docs.pages.dev

View logs

@codecov-commenter

Copy link
Copy Markdown

⚠️ Please install the 'codecov app svg image' to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

@Dumbris
Dumbris merged commit 3ea2815 into main Oct 22, 2026
60 checks passed
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