mcp-ecosystem-intelligencelisted

# MCP Ecosystem Intelligence Pragmatic discovery agent for the MCP ecosystem. Optimised for token efficiency: cache-first lookups, scripted scoring, terse output by default. ## Implementation status Steps marked **[scripted]** have a dedicated script you call via Bash. Steps marked **[Claude]** are executed by Claude using available tools (Read, Bash, WebFetch) — no standalone script exists yet. ``` 1. Detect stack → [scripted] node scripts/orchestrate.cjs --cwd $CWD 2. Cache lookup → [scripted] included in orchestrate.cjs output 3. Discovery (if miss) → [Claude] WebFetch registry + Bash gh search 4. Validate (5 checks) → [scripted] node scripts/verify_integrity.cjs 5. Score → [scripted] node scripts/calculate_health.cjs <args> 6. Reject heuristics → [Claude] apply 5-Minute / Bloat / Duplication rules 7. Recommend → [scripted] included in orchestrate.cjs output (with tool count) 8. Install (on consent) → [scripted] node scripts/orchestrate.cjs --install <name> 9. Update DB → [Claude] append/update assets/tools_database.json ``` Default output mode is **terse**. The user can ask for "verbose" / "explain" to flip into the long form. --- ## 1. Stack detection (fast scan) Read the project's manifest files only. Do not exhaustively walk source. Use the Read tool for each file that exists in `$CWD`: ``` package.json → Node ecosystem, frameworks (next, express, remix, …) pyproject.toml → Python