MCP Catalog Challenge: what does your model actually see?

[labmimors]

MCP Catalog Challenge: what does your model actually see?

MCP Lens now has a local-only calculator for one narrow question: how many tool schemas, and how many canonical UTF-8 JSON bytes, are visible to your model?

Try it here: https://labmimors.github.io/dsh-mcp-lens/

The page runs entirely in your browser. It has no upload API or external scripts. Please still remove server names, internal descriptions, credentials, URLs, and any other private metadata before pasting a catalog.

If you are willing to share a sanitized result, reply with:

Client / harness:
Number of MCP servers:
Model-visible tool schemas:
Canonical tool JSON bytes:
Lens comparison shown by the calculator:
Workload shape: fixed few tools / long tail / mixed
What you want measured next:

You can also attach the PNG generated by the calculator. The exported card contains only the summary numbers and claim boundary, not the pasted schemas.

What would be especially useful:

• multilingual or CJK tool names;
• near-duplicate tools across servers;
• a real search miss, with sanitized name/description and the query you expected to work;
• small catalogs where the extra search step is not worthwhile;
• large catalogs where request size becomes visible in usage or traces.

Please do not post API keys, auth headers, signed URLs, private server names, raw production schemas, or customer data.

The calculator measures JSON.stringify(tools) UTF-8 bytes. It does not claim token, billing, latency, or task-quality savings. The separate DeepSeek pilot and its tradeoffs are documented here:
https://github.com/labmimors/dsh-mcp-lens/blob/main/docs/LIVE_DEEPSEEK_PILOT.md

I will turn reproducible search misses into regression cases and keep any public comparison protocol visible in issue #9.

[labmimors]

rc.8 is now public: reviewed plugin release, retrieval report, and install path

Plugin release: https://github.com/labmimors/dsh-mcp-lens/releases/tag/v0.1.0-rc.8
Prebuilt plugin tarball: https://github.com/labmimors/dsh-mcp-lens/releases/download/v0.1.0-rc.8/dsh-mcp-lens-0.1.0-rc.8.tgz
Frozen retrieval report: https://github.com/labmimors/dsh-mcp-lens/blob/v0.1.0-rc.8/docs/RETRIEVAL_EVALUATION.md
Package SHA-256: a930b5166ffe1cf1de4032d69289de935c444c94cf01b2a5ca5ad58949b91fa0
Benchmark SHA-256: 2547d759983fa8bbb6a33b53ebee55c8c088f718bc52b37900b9c252b54413cd

Install note: download and verify the asset before passing its local path to dsh plugin add. Some pnpm versions reject a redirected GitHub asset URL with ERR_PNPM_MISSING_TARBALL_INTEGRITY; the current README contains the verified command sequence.

The calculator still measures locally and shares aggregate metrics only. The URL fragment contains bounded numeric tool and byte fields plus an accidental-edit checksum; it never contains the schemas you pasted.
Calculator: https://labmimors.github.io/dsh-mcp-lens/

The dependency-free MCP Lens Schema Audit Action is unchanged in rc.8, so the reviewed reference stays on rc.7:

• uses: labmimors/dsh-mcp-lens@v0.1.0-rc.7
  with:
  tools-file: artifacts/request-header.json
  max-tools: 100
  max-schema-bytes: 65536

If you try the calculator or the plugin, reply with only aggregate counts, your client or harness, and the workload shape. Please keep schemas, internal names, and credentials private.