DaveGold/mcp-metadata-demo
Demo MCP server + MCP apps, and the evals that tested its metadata: what actually reaches the model, what changes answers, and a skill that applies it to new and existing servers. Companion to the paper 'The Missing Layer'. Uses Dutch open
field: AI / ML
ARCHETYPE
MCP Hoarder
Six or more MCP servers wired in. Every tool, always in context.
CONTEXT TAX · EVERY TURN
up to ~43.6k tokens
Claude Code loads MCP tools on demand, so the real cost is usually lower.
Context hog · median rig: 2.3k · breakdown
Copy this rig
# review before running: this installs third-party code $ claude mcp add --transport http metadata-demo-cloud https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcp $ claude mcp add --transport http eval-thin https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpMinimal $ claude mcp add --transport http eval-schema https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpSchema $ claude mcp add --transport http eval-inline https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInline $ claude mcp add --transport http eval-inline-recipe https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInlineRecipe $ claude mcp add --transport http eval-words https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpWords $ claude mcp add --transport http eval-words-recipe https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpWordsRecipe $ claude mcp add --transport http eval-rich https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcp $ claude mcp add --transport http eval-opaque https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpOpaque $ claude mcp add --transport http eval-opaque-words https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpOpaqueWords $ claude mcp add --transport http eval-inline-conditional https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInlineConditional $ claude mcp add --transport http eval-inline-fact https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInlineFact $ claude mcp add --transport http eval-inline-instruction https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInlineInstruction $ claude mcp add --transport http eval-inline-oneline https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInlineOneline $ claude mcp add --transport http eval-g-recipe https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpGuidanceRecipe $ claude mcp add --transport http eval-best https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpBest $ npx degit DaveGold/mcp-metadata-demo/.claude ./rig-mcp-metadata-demo # inspect, then merge into .claude/
MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.
$ claude mcp add --transport http metadata-demo-cloud https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcp $ claude mcp add --transport http eval-thin https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpMinimal $ claude mcp add --transport http eval-schema https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpSchema $ claude mcp add --transport http eval-inline https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInline $ claude mcp add --transport http eval-inline-recipe https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInlineRecipe $ claude mcp add --transport http eval-words https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpWords $ claude mcp add --transport http eval-words-recipe https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpWordsRecipe $ claude mcp add --transport http eval-rich https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcp $ claude mcp add --transport http eval-opaque https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpOpaque $ claude mcp add --transport http eval-opaque-words https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpOpaqueWords $ claude mcp add --transport http eval-inline-conditional https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInlineConditional $ claude mcp add --transport http eval-inline-fact https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInlineFact $ claude mcp add --transport http eval-inline-instruction https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInlineInstruction $ claude mcp add --transport http eval-inline-oneline https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpInlineOneline $ claude mcp add --transport http eval-g-recipe https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpGuidanceRecipe $ claude mcp add --transport http eval-best https://europe-west4-mcp-metadata-demo.cloudfunctions.net/mcpBest
$ npx degit DaveGold/mcp-metadata-demo/.agents/skills/mcp-compat-check .claude/skills/mcp-compat-check $ npx degit DaveGold/mcp-metadata-demo/.agents/skills/rich-domain-mcp-server .claude/skills/rich-domain-mcp-server $ npx degit DaveGold/mcp-metadata-demo/.agents/skills/run-eval .claude/skills/run-eval
$ curl -fsSL --create-dirs -o .claude/agents/eval-best.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-best.md $ curl -fsSL --create-dirs -o .claude/agents/eval-guidance-recipe.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-guidance-recipe.md $ curl -fsSL --create-dirs -o .claude/agents/eval-inline-conditional.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-inline-conditional.md $ curl -fsSL --create-dirs -o .claude/agents/eval-inline-fact.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-inline-fact.md $ curl -fsSL --create-dirs -o .claude/agents/eval-inline-instruction.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-inline-instruction.md $ curl -fsSL --create-dirs -o .claude/agents/eval-inline-oneline.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-inline-oneline.md $ curl -fsSL --create-dirs -o .claude/agents/eval-inline-recipe.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-inline-recipe.md $ curl -fsSL --create-dirs -o .claude/agents/eval-inline.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-inline.md $ curl -fsSL --create-dirs -o .claude/agents/eval-opaque-words.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-opaque-words.md $ curl -fsSL --create-dirs -o .claude/agents/eval-opaque.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-opaque.md $ curl -fsSL --create-dirs -o .claude/agents/eval-rich.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-rich.md $ curl -fsSL --create-dirs -o .claude/agents/eval-schema.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-schema.md $ curl -fsSL --create-dirs -o .claude/agents/eval-thin.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-thin.md $ curl -fsSL --create-dirs -o .claude/agents/eval-words-recipe.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-words-recipe.md $ curl -fsSL --create-dirs -o .claude/agents/eval-words.md https://raw.githubusercontent.com/DaveGold/mcp-metadata-demo/main/.claude/agents/eval-words.md
This rig commits no guardrails. Here is the community baseline instead — the deny/ask rules most often found across all 16.2k rigs:
{
"permissions": {
"deny": [
"Read(./.env)",
"Read(./.env.*)",
"Read(**/.env)",
"Read(~/.ssh/**)",
"Bash(git push --force:*)",
"Read(**/*.pem)",
"Read(.env)",
"Read(~/.aws/**)",
"Read(**/*.key)",
"Read(**/.env.*)",
"Bash(git push --force*)",
"Bash(rm -rf:*)",
"Bash(rm -rf *)",
"Bash(sudo *)",
"Bash(sudo:*)",
"Bash(rm -rf /)",
"Bash(git push -f:*)",
"Read(.env.*)",
"Bash(git reset --hard*)",
"Bash(rm -rf /*)"
],
"ask": [
"Bash(git push:*)",
"Bash(git push *)",
"Bash(gh pr merge *)",
"Bash(git commit:*)",
"Bash(rm *)",
"Bash(gh pr merge:*)",
"Bash(gh pr create *)",
"Bash(git reset --hard:*)",
"Bash(npm publish:*)",
"Bash(git rebase *)"
]
}
} MCP servers (17)
Skills (3)
Subagents (15)
| eval-best | Eval arm best (the reference implementation built with the rich-domain-mcp-server skill; Q19, app tools rebuilt in Q22, chart schema guided in Q25). Answers one building-data question using ONLY the e |
| eval-guidance-recipe | Eval arm guidance-recipe (Q8 — the DERIVED FIGURES recipe, returned by a no-argument guidance call). Answers one building-data question using ONLY the eval-g-recipe MCP server. Not for general use — s |
| eval-inline-conditional | Eval arm inline-conditional (Q2 — INTERPRETATION pruned to the record: matching berekeningstype branch plus notes for populated fields only). Answers one building-data question using ONLY the eval-inl |
| eval-inline-fact | Eval arm inline-fact (Q4 — the CALCULATED vs MEASURED line reduced to its FACT half: what the quantities are, no guidance on what to output). Answers one building-data question using ONLY the eval-inl |
| eval-inline-instruction | Eval arm inline-instruction (Q4 — the CALCULATED vs MEASURED line reduced to its INSTRUCTION half: what to output, with no reason given). Answers one building-data question using ONLY the eval-inline- |
| eval-inline-oneline | Eval arm inline-oneline (Q14 — words' description byte for byte, PLUS one sliced INTERPRETATION line in the response). Answers one building-data question using ONLY the eval-inline-oneline MCP server. |
| eval-inline-recipe | Eval arm inline-recipe (Q1b — the DERIVED FIGURES recipe, shipped by one of two channels). Answers one building-data question using ONLY the eval-inline-recipe MCP server. Not for general use — spawne |
| eval-inline | Eval arm inline (schema's description, INTERPRETATION prose in the RESPONSE). Answers one building-data question using ONLY the eval-inline MCP server. Not for general use — spawned by the run-eval sk |
| eval-opaque-words | Eval arm opaque-words. Answers one building-data question using ONLY the eval-opaque-words MCP server. Not for general use — spawned by the run-eval skill. |
| eval-opaque | Eval arm opaque. Answers one building-data question using ONLY the eval-opaque MCP server. Not for general use — spawned by the run-eval skill. |
| eval-rich | Eval arm rich. Answers one building-data question using ONLY the eval-rich MCP server. Not for general use — spawned by the run-eval skill. |
| eval-schema | Eval arm schema (minimal + typed schemas, no prose). Answers one building-data question using ONLY the eval-schema MCP server. Not for general use — spawned by the run-eval skill. |
| eval-thin | Eval arm thin. Answers one building-data question using ONLY the eval-thin MCP server. Not for general use — spawned by the run-eval skill. |
| eval-words-recipe | Eval arm words-recipe (Q1b — the DERIVED FIGURES recipe, shipped by one of two channels). Answers one building-data question using ONLY the eval-words-recipe MCP server. Not for general use — spawned |
| eval-words | Eval arm words. Answers one building-data question using ONLY the eval-words MCP server. Not for general use — spawned by the run-eval skill. |
Similar rigs
hishamkaram/markdown-memory
MCP server that hands coding agents the one Markdown section that answers the question, not the whole file. Local hybrid search: BM25 keywords + ONNX embeddings in SQLite. No API keys, no network.
Pragmatist 8.4k tok · · Q3
sloprail/sloprail
Deterministic project rules for your coding agent. They kick in when it touches a file or runs a command, and hold until met: files only where you allow, right skills loaded first, work proven by diff or logs. Not a suggestion in CLAUDE.md
Pragmatist 624 tok · · Q15
anthropics/knowledge-work-plugins
Open source repository of plugins primarily intended for knowledge workers to use in Claude Cowork
MCP Hoarder 118.2k tok · · Q0
ParadevAI/Paradev
Paradev is an open-source Codex App and Conductor alternative. Orchestrate an army coding agents in parallel with your own subscriptions.
MCP Hoarder 30.4k tok · · Q23