plurigrid/asi / ies/music-topos/.agents/skills/mcp-tripartite
mcp-tripartite
Verified| File | Kind | Size |
|---|---|---|
| MCP_INVENTORY.md | markdown | 3.5 KB |
| MCP_OPTIMAL_TRANSITIONS.md | markdown | 6.7 KB |
| MCP_USAGE_ASSESSMENT.md | markdown | 4.5 KB |
| SKILL.md | skill md | 7.8 KB |
MCP_USAGE_ASSESSMENT.md
View on GitHub# MCP Usage Assessment: Historical Analysis
## Discovery Sources Analyzed
| Source | Records | MCP Mentions |
|--------|---------|--------------|
| `~/.codex/history.jsonl` | 791 | 110+ |
| `~/.codex/history.duckdb` | 18 events | thread closures |
| `.crush.json` | 5 MCPs configured |
| `~/.codex/config.toml` | 7 MCPs configured |
---
## MCP Usage Frequency (from Codex History)
| MCP Server | Mentions | Trit | Energy | Usage Pattern |
|------------|----------|------|--------|---------------|
| **gay** | 223 | 0 | LOW | Core system, color generation |
| **exa** | 128 | +1 | HIGH | Search, research |
| **firecrawl** | 87 | +1 | HIGH | Web scraping |
| **unison** | 50 | 0 | MEDIUM | Content-addressed code |
| **tree-sitter** | 43 | -1 | LOW | Code analysis |
| **marginalia** | 26 | +1 | MEDIUM | Indie search |
| **radare2** | 24 | -1 | MEDIUM | Binary analysis |
| **huggingface** | 20 | 0 | MEDIUM | Model/paper search |
| **babashka** | 5 | 0 | LOW | Clojure scripting |
---
## Usage Distribution by Trit
```
MINUS (-1): 67 mentions (tree-sitter + radare2) = 11%
ERGODIC (0): 298 mentions (gay + unison + hf + bb) = 49%
PLUS (+1): 241 mentions (exa + firecrawl + marg) = 40%
Total: 606 mentions
GF(3) check: 67 + 298 + 241 = 606 ≡ 0 (mod 3) ✓
```
---
## Optimal Transition Patterns (Observed)
### Pattern 1: Research Flow (Most Common)
```
exa (+1) → gay (0) → tree-sitter (-1)
```
- **Frequency**: High (exa + gay dominates)
- **Energy**: 5 + 1 + 1 = 7
- **GF(3)**: 1 + 0 + (-1) = 0 ✓
### Pattern 2: Code Analysis Flow
```
tree-sitter (-1) → gay (0) → marginalia (+1)
```
- **Frequency**: Medium
- **Energy**: 1 + 1 + 3 = 5 (optimal!)
- **GF(3)**: -1 + 0 + 1 = 0 ✓
### Pattern 3: Deep Web Scrape
```
firecrawl (+1) → unison (0) → radare2 (-1)
```
- **Frequency**: Low but powerful
- **Energy**: 5 + 3 + 3 = 11
- **GF(3)**: 1 + 0 + (-1) = 0 ✓
---
## Energy Efficiency Analysis
### Current Usage
```
Total Energy Spent:
gay(223×1) + exa(128×5) + firecrawl(87×5) + unison(50×3) +
tree-sitter(43×1) + marginalia(26×3) + radare2(24×3) +
huggingface(20×3) + babashka(5×1)
= 223 + 640 + 435 + 150 + 43 + 78 + 72 + 60 + 5 = 1706 energy units
```
### Information Density
- **gay**: Highest (deterministic, instant, always correct)
- **tree-sitter**: High (structural analysis)
- **exa/firecrawl**: Medium (web content variable)
### Optimization Opportunities
1. **Replace exa with marginalia** when searching for obscure/indie content
- Energy savings: 5 → 3 per call
- Potential savings: ~250 energy units
2. **Use gay for seed chaining** before other calls
- Provides deterministic ordering
- Zero additional latency
3. **Cache firecrawl results** aggressively
- TTL: 24 hours
- Potential savings: ~200 energy units
---
## Recommended Triads by Use Case
### Use Case: Research/Discovery
```
exa (+1) → huggingface (0) → tree-sitter (-1)
```
- Find papers → Find models → Analyze code
- Energy: 9, Latency: 1550ms
### Use Case: Code Understanding
```
tree-sitter (-1) → gay (0) → marginalia (+1)
```
- Parse → Color → Find docs
- Energy: 5, Latency: 560ms (OPTIMAL)
### Use Case: Binary/Security
```
radare2 (-1) → unison (0) → exa (+1)
```
- Disassemble → Hash-address → Search CVEs
- Energy: 11, Latency: 1500ms
### Use Case: Content Transformation
```
firecrawl (+1) → babashka (0) → tree-sitter (-1)
```
- Scrape → Transform → Validate
- Energy: 7, Latency: 2150ms
---
## Self-Play Assessment Loop
```
Current State → Predict Optimal MCP → Execute → Measure → Refine
Loop invariant: GF(3) = 0 at each step
Convergence: Energy/Information ratio minimized
```
### Metrics to Track
| Metric | Current | Target |
|--------|---------|--------|
| Avg Energy/Call | 2.8 | < 2.0 |
| GF(3) Violations | 0 | 0 |
| Cache Hit Rate | N/A | > 70% |
| Latency p95 | ~2s | < 1s |
---
## Commands
```bash
# Full assessment
just mcp-tracker
# Usage by server
just mcp-triads-for gay
# Optimal for use case
just mcp-optimal-triad
# Historical analysis
grep -c "firecrawl\|exa\|gay" ~/.codex/history.jsonl
```
---
## Conclusion
The current MCP usage is **heavily weighted toward ERGODIC (0)** servers, particularly `gay` for deterministic coloring. The high usage of `exa` (+1) creates an imbalance that should be compensated with more `tree-sitter` (-1) calls for analysis.
**Key Insight**: The `gay → tree-sitter → marginalia` triad is optimal for most code understanding tasks with only 5 energy units and 560ms latency.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |