ProxyRank← Run another audit

https://github.com/github/github-mcp-server

78
out of 75 pts available
Discoverable

Selection Probability Lift

2.3×

vs baseline orchestrator selection

⚠ Provisional estimate — calibrated with real orchestrator data in Phase 2.

Full 100 pts available in Phase 2 with live probe enabled. Currently 75 pts max — Reliability probe adds 18 pts when activated.

Layer Radar

All layers normalized to 0–100 for visual comparison.

Audit Details

Manifest URL
https://github.com/github/github-mcp-server
Mode
MCP
Tools in manifest
16
Total manifest tokens
740
Avg tokens / tool
46
Base score (pre-penalty)
78
Penalty applied
No

Layer Breakdown

Orchestrator Simulation

How 4 major AI orchestrators evaluate your agent

4/4

would select

75
Selects

LangGraph

Graph-based agent orchestration

↑Schema layer is strong — LangGraph can reliably pass typed parameters between graph nodes.
↓Reliability data unavailable — LangGraph retry logic cannot be tuned for this tool.

Fix: Add async:true and document latency expectations.

74
Selects

CrewAI

Role-based multi-agent delegation

↑Semantic layer is strong — CrewAI agents can infer the tool's role from its description.
↓Schema coverage is thin — structured output parsing may fail in some crew workflows.

Fix: Add required: [] array to parameter schema.

77
Selects

OpenAI Assistants

Function calling via Assistants API

↑Schema is well-defined — OpenAI function calling can serialize inputs correctly.
↓Reliability unknown — Assistants API cannot estimate retry cost for this function.

Fix: Add output schema with type and properties.

74
Selects

AutoGen

Microsoft multi-agent conversations

↑Description is clear enough for AutoGen agents to route conversation turns.
↓Governance signals absent — AutoGen cannot assess tool trustworthiness in multi-agent pipelines.

Fix: Add signing metadata to pass AutoGen's trust filter.

⚠ Simulated thresholds — based on documented selection criteria, not live orchestrator calls.

vs. Industry Leaders

How your agent compares to well-known MCP implementations (all layers normalized to 0–100).

LayerYour AgentStripe Agent ToolkitGitHub MCP ServerSlack MCP Server
Overall78817872
Semantic83868377
Schema80878073
Reliability28242828
Governance60706060