[skakri]

Problem

Clients that negotiate MCP protocol version 2026-07-28 require ttlMs and cacheScope on list results (SEP-2549) and reject a tools/list response without them:

Invalid result for tools/list:
  ttlMs: Invalid input: expected number, received undefined
  cacheScope: Invalid option: expected one of "public"|"private"

Because the rejection happens at tools/list, the server connects with no tools at all — every session against a strict 2026-07-28 client is bricked.

Fix

The hand-written ServerHandler::list_tools now emits immediately-stale public cache hints (ttlMs: 0, cacheScope: "public") when the negotiated protocol version is 2026-07-28 or newer, and keeps the pre-SEP-2549 wire shape for legacy peers that may reject unknown fields. This mirrors what rmcp's #[tool_handler] macro does since 3.1.1 (modelcontextprotocol/rust-sdk#1120) — the macro fix doesn't reach hand-written handlers, so the gate is applied at our handler.

ttlMs: 0 deliberately opts out of client-side caching: the catalog is static per binary, not per server identity — hot-upgrade (SIGUSR1 exec) and version bumps can change it with no client-visible signal, and the only fetch it would save is one local stdio round trip per session. A real TTL should only land together with tools/listChanged wiring.

Also bumps rmcp 3.0.1 → 3.1.2.

Verification

• Two new tests drive tools/list through a real in-process client/server pair (rmcp client feature as a dev-dependency, scoped out of the production feature set by resolver 3), asserting the hints are present for a 2026-07-28 peer and absent for a legacy peer; the positive test fails if the gate is reverted.
• Full rag-rat-mcp suite green under nextest and cargo test; clippy --no-default-features and default-features legs clean under -D warnings; nightly rustfmt applied.

[github-actions]

SCIP oracle — resolution report

Heuristic→compiler edge resolution per corpus. Δ compares resolved-after to the main baseline (only when the corpus profile + tool version match).

corpus	tool	edges	resolved (heuristic → compiler)	precision	recall	monikers	Δ vs main
c-libuv	scip-clang	12894	37.9% → 48.6%	90.2%	55.3%	679	— (no baseline)
cpp-yaml	scip-clang	9423	43.2% → 72.9%	82.5%	40.9%	1056	— (no baseline)
py-django	scip-python	12937	28.8% → 73.5%	96.7%	84.1%	3701	— (no baseline)
py-rich	scip-python	8383	27.6% → 92.3%	95.0%	81.3%	1196	— (no baseline)
rust-time	rust-analyzer	11040	37.6% → 59.1%	72.7%	83.6%	574	— (no baseline)
ts-rxjs	scip-typescript	7875	39.6% → 94.9%	87.3%	77.4%	631	— (no baseline)

_{resolved = Exact/Syntactic + compiler upgrades + resolved-external, over edge candidates with a callee range. precision/recall are the oracle eval metrics.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!

[github-actions]

Bencher Report

Project	rag-rat
Branch	chore/rmcp-bump
Testbed	ubuntu-latest

⚠️ WARNING: No Threshold found!

Without a Threshold, no Alerts will ever be generated.

• Estimated Cycles (cycles)
• Instructions (instructions)
• RAM Hits (hits)
• Total read+write (reads/writes)
• LL Hits (hits)
• L1 Hits (hits)

Click here to create a new Threshold
For more information, see the Threshold documentation.
To only post results if a Threshold exists, set the --ci-only-thresholds flag.

Click to view all benchmark results

Benchmark	Estimated Cycles	cycles x 1e6	Instructions	instructions x 1e6	L1 Hits	hits x 1e6	LL Hits	hits x 1e6	RAM Hits	hits x 1e3	Total read+write	reads/writes x 1e6
rag_pipeline::pipeline::index cargo_resolver:resolver_config()	📈 view plot ⚠️ NO THRESHOLD	1,952.53 x 1e6	📈 view plot ⚠️ NO THRESHOLD	1,233.57 x 1e6	📈 view plot ⚠️ NO THRESHOLD	1,714.18 x 1e6	📈 view plot ⚠️ NO THRESHOLD	38.64 x 1e6	📈 view plot ⚠️ NO THRESHOLD	1,290.64 x 1e3	📈 view plot ⚠️ NO THRESHOLD	1,754.11 x 1e6
rag_pipeline::pipeline::query_cold cargo_resolver:resolver_built_config()	📈 view plot ⚠️ NO THRESHOLD	340.60 x 1e6	📈 view plot ⚠️ NO THRESHOLD	226.81 x 1e6	📈 view plot ⚠️ NO THRESHOLD	323.73 x 1e6	📈 view plot ⚠️ NO THRESHOLD	3.15 x 1e6	📈 view plot ⚠️ NO THRESHOLD	31.66 x 1e3	📈 view plot ⚠️ NO THRESHOLD	326.92 x 1e6
rag_pipeline::pipeline::query_warm cargo_resolver:resolver_index()	📈 view plot ⚠️ NO THRESHOLD	314.11 x 1e6	📈 view plot ⚠️ NO THRESHOLD	208.79 x 1e6	📈 view plot ⚠️ NO THRESHOLD	298.39 x 1e6	📈 view plot ⚠️ NO THRESHOLD	2.91 x 1e6	📈 view plot ⚠️ NO THRESHOLD	32.98 x 1e3	📈 view plot ⚠️ NO THRESHOLD	301.34 x 1e6

🐰 View full continuous benchmarking report in Bencher

[github-actions]

Bencher Report

Project	rag-rat
Branch	chore/rmcp-bump
Testbed	ubuntu-latest

Click to view all benchmark results

Benchmark	Latency	Benchmark Result seconds (s) (Result Δ%)	Upper Boundary seconds (s) (Limit %)
index_time/full_rebuild_cargo	📈 view plot 🚷 view threshold	5.93 s (+6.23%) Baseline: 5.58 s	6.32 s (93.83%)

🐰 View full continuous benchmarking report in Bencher