Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add gleanwork/mcp-server-tester --skill optimize-mcp-tool-metadatagit clone --depth 1 https://github.com/gleanwork/mcp-server-testerWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/gleanwork/mcp-server-tester/optimize-mcp-tool-metadata)<a href="https://agentmods.dev/skills/gleanwork/mcp-server-tester/optimize-mcp-tool-metadata"><img src="https://agentmods.dev/badge/skills/gleanwork/mcp-server-tester/optimize-mcp-tool-metadata/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/gleanwork/mcp-server-tester/optimize-mcp-tool-metadata"><img src="https://agentmods.dev/badge/skills/gleanwork/mcp-server-tester/optimize-mcp-tool-metadata.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00106 | $0.02826 |
| Opus 5.5 | $0.00042 | $0.01130 |
| Sonnet 5.5 | $0.00021 | $0.00565 |
| Haiku 4.5 | $0.00011 | $0.00283 |
Grade A, and why
optimize-mcp-tool-metadata scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 22d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 232 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Optimize MCP tool metadata
A tool optimization runs tool-metadata variants against a baseline and recommends applying one or none. Each variant shows the client different tool names, descriptions or input schemas; the cases, servers, client and model stay the same. You decide which rewrites to try. MST runs the trials, compares each variant with the baseline case by case, guards the cases that work today, and returns a proposal.
Rules
- Don't edit the dataset. It defines what good behaviour is. Variants are data passed to the run.
- Don't edit the server source unless the user asks. Deliver the proposal and the text to paste.
- Respect the regression guard. A variant that fixes two cases and breaks one isn't a win. Leave
allowRegressionsoff unless the user accepts the trade. - Mind cost. Every variant is a full run: cases × trials × a model conversation each. Try a few variants built on evidence, not many guesses.
What a variant can change
Tool metadata changes what the client sees, not what the server accepts. The model forms calls against the variant's metadata, and the calls run on the real server.
- Change freely: a tool's
description; descriptive text in itsinputSchema(property descriptions, enum documentation, examples, format hints). - Rename with care:
namegives the tool a new name. Calls reach the original tool and are recorded under the original name, so assertions don't change. The server keeps its own name until someone changes it. - Never change: parameter names, types,
required, or schema structure. The server still validates real calls, so the run would measure rejected calls, not discoverability. - Out of scope: tool behaviour, response shapes, auth or a different server build. Compare those as variants with different
servers(seewrite-mcp-eval).
Prerequisites
- A dataset of cases with
toolsTriggeredassertions (thewrite-mcp-evalskill). Several phrasings per tool, and inputs that should not trigger it. - Cases that work today and must keep working tagged
regression. Without the tag, MST runs the baseline one extra time to find them. - Optionally, a few cases tagged
held-out. They don't count toward ranking and your proposals never see them, so their result shows whether the winner generalizes. - An API key for the model (
ANTHROPIC_API_KEYforclaude-*models on themstclient).
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 22d ago Changed · +63 lines · +29 tokens per session 8c3382743ff8
- 23d ago Changed 27477252dff6
- 25d ago Changed e01b11aca462
- 28d ago Changed · -2 lines 7577f18920d1
- 2mo ago First seen · 171 lines · 77 tokens per session scan A c91076780374
optimize-mcp-tool-metadata is a skill published in the GitHub repository gleanwork/mcp-server-tester (19 stars, last pushed today), licensed MIT. It adds 106 tokens to every session and 2,826 once invoked, about $0.0004 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
local-ai-agents
Build local-first AI agents that run entirely on a developer workstation with Microsoft Foundry Local and Qwen function-calling models. Covers Small Language Models (SLMs), the OpenAI-compatible local endpoint, sandboxed local tools, local RAG with Chroma, local MCP servers, hybrid cloud/local routing, and the…
systematic-debugging
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.
next-cache-components-adoption
Turn on Cache Components in a Next.js app and resolve the blocking routes it surfaces. Use when the user wants to enable, adopt, or migrate to Cache Components, flip the cacheComponents flag, work through a flood of blocking-prerender / instant validation errors, run the cache-components-instant-false codemod, or…
chat-pet-sprite-creation
Use when creating or changing VS Code chat pet sprite art, sprite sheets, state animations, eye treatments, Stable/Insiders variants, or pet transitions under src/vs/workbench/contrib/chat/browser/widget/media/chatPet.
cpu-profile-analysis
Analyze V8/Chrome CPU profiles (.cpuprofile) and DevTools trace files (Trace-.json). Use when: profiling performance, investigating slow functions, comparing code paths, finding bottlenecks, analyzing timeToRequest, understanding call trees from sampling profiler data, analyzing layout/paint/rendering, investigating…
insight-error-page
Write or audit an insight-kind error page for the Next.js dev overlay. Use when creating a new errors/ .mdx page, auditing an existing one, or checking that a page matches the framework fix cards. Covers page structure, title alignment, FixCard cards with Copy prompt button, code snippets, terminology verification…