Repository navigation
feat(connections): register Quo, launch-gated - #30668
Conversation
Sensitive Change ReviewSensitive files changed:
How to clear this check:
Only the author's comment counts this way. A reviewer can instead |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Nothing to check. One hidden catalogue entry plus the exact consent-host pair its Connect flow needs; no default changes while What this change shipsInventory (4 items) — 4 justifiedIntent: Quo users get a curated, read-scoped Connect for Quo's hosted MCP server, kept hidden until the manual launch-gate check passes. ADDITION.
[FIRST-PRINCIPLES-REVIEWED] 0e2a5ca Description read[DESCRIPTION-READ] 9cf1c1577e7e2bac0a447f9525159f31149a7faec5e600881d1b3d833cbc9dfa That sha256 names the PR description this verdict read. Recompute it with |
UX Review (Opus 5.5) — ⏭️ skippedRevision |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: this adds one hidden, launch-gated registry entry and one exact consent-host pair, takes nothing away, and follows the TickTick precedent. Notes
[DESIGN-REVIEWED] 0e2a5ca Description read[DESCRIPTION-READ] 6629f02a6977c27f65f1fdac5312b0f196ce7531fd584ff4878186e8816d56b3 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Security Scope Review -- ✅ PASSWhich legitimate operations does Verdict basis: adjudicated, zero confirmed regressions. Security scope review -- adjudicated candidates
No candidate flipped from allowed to refused. Scope-Verdict: PASS Reason: Nothing in this diff makes a check stricter. Its only security-relevant hunk adds one exact TIGHTENING: none. The one security edit loosens a check: ADJUDICATED: 14 candidates, 0 confirmed newly-refused PLATFORM COVERAGE: The diff adds no assumption about paths, separators, executables or environment variables. The new pair is a host-and-path string compared exactly, the same on every platform.
No platform assumption in the diff is left without a candidate. UNADJUDICATED: The guard is the ACP banner-safety OAuth carve-out: [SCOPE-REVIEWED] 0e2a5ca |
04ad1d6 to
adb33cc
Compare
Add Quo's hosted MCP server (https://mcp.quo.com/mcp) to the Connections registry as a tier-2 entry with launch_gate_passed=false, so the Connect grid does not change until the manual launch-gate check passes. Quo is listed by both connector directories and publishes RFC 9728 resource metadata naming an RFC 8414 issuer with dynamic client registration and PKCE, so the L0 probe passes and no operator app is needed. The entry asks only for Quo's read scopes. Its gotcha copy names the tools that send texts, have Quo's agent call or text people, spend credits or file a billed carrier registration. Quo's consent endpoint joins the OAuth banner allowlist and the URL corpus; the user doc, the docs index and the spec roster count the new entry. Closes #19182
adb33cc to
0e2a5ca
Compare
|
Rebased: main moved in docs/system-specs/modules/, src/kiro_crew/docs/ and test/ (same-directory neighbours only; none of this PR's own files changed, and the patch is unchanged) since this head went green (7a09d86 -> 8fb88a6); CI re-running. The new head also replaces the Security Scope Review red on the old head: that lane's verdict was PASS, and the job failed afterwards on a GitHub API rate limit while reading its own earlier results, which a rerun of the same head cannot clear. [operator: bolichen97] |
Problem / Motivation
Goal: Quo users get a catalogue-governed Connect for Quo's hosted MCP server, hidden behind the launch gate until the manual check passes.
The Connections catalogue has no Quo entry. A user who wants an agent to read their Quo texts, calls and contacts has to add
https://mcp.quo.com/mcpby hand, and gets none of the catalogue's guarantees: no nightly OAuth metadata watch, no consent-banner allowlisting, no narrowed scopes and no copy saying which tools text or call real people.Why it matters
Quo (formerly OpenPhone) hosts its own remote MCP server, is listed in both the Claude and ChatGPT connector directories, and publishes standard OAuth discovery with dynamic client registration and PKCE, so it clears the first rungs of the admission path. Its tools can send texts, have Quo's AI agent place calls, and file a billed carrier registration, so a curated entry that asks for read access only and names those tools is worth more here than for most vendors.
Not a goal
launch_gate_passed: the Connect grid renders the same cards as main until a human walks the flow with a real Quo account.revoke_verified_notesayspending manual check.What changed (motivation → approach → change)
Quo joins
src/kiro_crew/connections/registry.jsonhidden, the same way TickTick did (#29289): tier 2,launch_gate_passed: false, categorycalendar-email(the registry's communication bucket; Claude's directory files Quo under Communication).https://mcp.quo.com/mcp(401 withresource_metadatapointing at its RFC 9728 document)https://mcp.quo.com, withregistration_endpoint(DCR) and PKCES256read:account,read:conversations,read:calls, out of eight inscopes_supportedlist-inboxes, a read-only tool with no required argumentsQuo is the rare vendor that publishes separate read and write scopes, so the entry asks only for the read ones and leaves the write half of the issue's title out on purpose. Quo does not document whether a read-only grant can still call its write tools, so the entry's
gotcha_copy(reference text, not rendered on the card) says to treat them as callable and names the ones that send texts and spend credits, the two that have Quo's Sona agent call or text a person on its own, the billed carrier registration, and the inbox-settings change that applies to live calls. The consent URLmcp.quo.com/authorizejoins_OAUTH_AUTHORIZATION_ENDPOINTSinsecurity/exfil_oauth.pyand the URL corpus as one exact host-and-path pair, which is rung 4 of the admission path. The user doc (src/kiro_crew/docs/connections.md,src/kiro_crew/docs/index.md) counts 38 providers, and the spec roster gains a Quo row.Backwards compatibility
Compatible: the registry gains one hidden entry and the banner allowlist one exact pair; nothing main accepts is rejected.
Removes nothing: every touched file only gains lines or a bumped count.
Tests
test/test_connections_registry.py: Quo is on the roster and launch-gated.test_quo_requests_only_read_scopes_and_names_the_tools_that_send_or_spendpins the endpoint, the three read scopes, and the seven send, call, billed and live-settings tools named in the copy.test/oauth_url_corpus.py: a Quo consent URL carrying the three scopes, which the banner allowlist must accept.Ran on this head (main moved in none of this PR's files since the previous head, so the patch is unchanged): the backend lane of
scripts/local-gate.py(59 related files, 2619 passed),mypyon the touched module, black/isort/flake8 on the touched files,check_comment_history.py,check_brand_name.py,check_file_line_cap.pyanddocs-lint.sh. On an earlier head, before the rebase past the Railway entry (#30662), which landed on the same lines: the three vitest files that import the real registry (141 passed),mypy src/kiro_crew, the flake-pattern ratchet and semgrep 1.78.0 on the diff. Determinism, on that earlier head: the new test passed 20 times at-n0, its file plus the URL-corpus and line-budget files passed 10 times at-n 4, and the registry and URL-corpus files passed in a shuffled order (seed 8658).Manual verification
Done:
python -m kiro_crew.connections.l0_probeon the previous head, whose registry is byte-identical to this one: 38 of 38 providers ok, Quo observedauthorization_server https://mcp.quo.com, dcr true, pkce true.tools/listwas run from this checkout.Still owed before the card goes visible (the launch-gate check, on a test gateway with a real Quo account):
revoke_page_url(Quo's MCP security page today) with what the check finds.list-contactsandget-contact(Quo has noread:contactsscope), and thatlist-inboxesanswers.Screenshots / video
N/A — the diff touches no frontend file, and the entry ships with
launch_gate_passed=false, so the Connect grid renders the same cards as main.Related Issues
Closes #19182
Split out of #7130. Prior art: #29289 (TickTick), the entry this one mirrors and which closed its issue the same way: the remaining manual launch-gate walk is tracked in the spec roster's gate column, like every other gated entry.
Sibling sweep
N/A — a
featPR, so there is no defect to sweep.Pattern harvest
N/A — a
featPR.Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)