Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →MCP registry steward for Claude and Codex. Inventories available MCP servers/tools, recommends safe registration, and maintains project MCP usage records without inventing unavailable capabilities.
.claude/skills/30eggis-support-support-mcp-registry-steward/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 75% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 344% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 128% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 11% | 0% |
You manage Model Context Protocol capability discovery, registration guidance, and maintenance for Claude and Codex runtimes.
Help COO and the responsible CXX decide whether MCP tools should be used for a mission, and keep the decision auditable.
None with the exact reason..claude MCP configuration applies to Codex.For every proposed MCP registration, document:
Write or update a worker report under:
.harness/documents/{mission}/coo/workers/support-support-mcp-registry-steward.md
Required sections:
## MCP Capability Inventory## Claude Runtime## Codex Runtime## Registration Candidates## Risk Review## Maintenance Plan## Recommended Mission Use## Implementation NotesUse None for empty sections. Include exact evidence paths or tool names for every claim.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 16,942 | 4,278 | -75% | 1 | 1 | 0% | 3,055 | 915 | -70% | 0 | 0 | — |
case-07 | pass→pass | 14,159 | 12,456 | -12% | 1 | 1 | 0% | 2,704 | 2,835 | +5% | 0 | 0 | — |
case-02 | fail→fail | 35,683 | 3,270 | -91% | 1 | 1 | 0% | 5,451 | 873 | -84% | 0 | 0 | — |
case-03 | fail→fail | 10,765 | 6,198 | -42% | 1 | 1 | 0% | 2,369 | 952 | -60% | 0 | 0 | — |
case-04 | pass→pass | 15,201 | 20,765 | +37% | 1 | 1 | 0% | 3,430 | 4,520 | +32% | 0 | 0 | — |
case-05 | pass→fail | 8,186 | 16,900 | +106% | 1 | 1 | 0% | 1,754 | 4,472 | +155% | 0 | 0 | — |
case-06 | pass→fail | 4,255 | 4,621 | +9% | 1 | 1 | 0% | 693 | 1,044 | +51% | 0 | 0 | — |
case-08 | pass→pass | 12,882 | 15,562 | +21% | 1 | 1 | 0% | 2,165 | 3,523 | +63% | 0 | 0 | — |
case-09 | fail→pass | 9,562 | 14,411 | +51% | 1 | 1 | 0% | 1,924 | 3,374 | +75% | 0 | 0 | — |
case-10 | pass→pass | 11,587 | 13,841 | +19% | 1 | 1 | 0% | 1,884 | 2,109 | +12% | 0 | 0 | — |
case-11 | pass→pass | 6,147 | 7,505 | +22% | 1 | 1 | 0% | 1,089 | 1,920 | +76% | 0 | 0 | — |
case-12 | fail→pass | 5,067 | 20,284 | +300% | 1 | 1 | 0% | 886 | 3,932 | +344% | 0 | 0 | — |
case-13 | fail→pass | 14,793 | 11,044 | -25% | 1 | 1 | 0% | 2,451 | 2,679 | +9% | 0 | 0 | — |
case-14 | fail→fail | 7,270 | 6,257 | -14% | 1 | 1 | 0% | 1,175 | 1,070 | -9% | 0 | 0 | — |
case-15 | fail→pass | 4,732 | 6,414 | +36% | 1 | 1 | 0% | 817 | 1,863 | +128% | 0 | 0 | — |
case-16 | fail→fail | 23,207 | 4,879 | -79% | 1 | 1 | 0% | 3,701 | 954 | -74% | 0 | 0 | — |
case-22 | pass→pass | 15,009 | 14,849 | -1% | 1 | 1 | 0% | 3,166 | 3,683 | +16% | 0 | 0 | — |
case-17 | pass→fail | 12,353 | 7,505 | -39% | 1 | 1 | 0% | 2,220 | 1,078 | -51% | 0 | 0 | — |
case-18 | pass→pass | 15,002 | 13,228 | -12% | 1 | 1 | 0% | 2,214 | 2,957 | +34% | 0 | 0 | — |
case-19 | fail→pass | 20,441 | 15,495 | -24% | 1 | 1 | 0% | 3,167 | 3,505 | +11% | 0 | 0 | — |
case-20 | fail→pass | 4,431 | 12,281 | +177% | 1 | 1 | 0% | 880 | 2,736 | +211% | 0 | 0 | — |
case-21 | pass→fail | 12,527 | 3,782 | -70% | 1 | 1 | 0% | 2,070 | 957 | -54% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 14 counted toward the lift figure. The other 8 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +9 percentage points is the difference between those two pass rates over the 14 comparable cases. 6 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.