[mcp-analysis] Structural Analysis - 2026-06-09

[github-actions[bot]]

Overview

Today's run probed 8 representative GitHub MCP tools across the available toolsets (~3,075 tokens total, avg usefulness 3.12/5 including two environmentally-unavailable tools). Among the 6 tools that actually returned data, the average rating is 3.83/5.

The schema story is unchanged and consistent with the 30-day trend: discovery/list tools are lean and context-efficient, while per-item detail tools are content-driven (full issue/PR bodies) or carry static-documentation bloat (list_code_scanning_alerts re-inlines the full CodeQL rule help per alert). search_repositories remains best-in-class; list_code_scanning_alerts remains the standout bloat case.

Key Metrics

Metric	Value
Tools probed	8 (6 returned data, 2 unavailable)
Total tokens (today)	~3,075
Avg usefulness (all)	3.12 / 5
Avg usefulness (available)	3.83 / 5
Best rated	search_repositories — 5/5
Most context-heavy	list_code_scanning_alerts — ~1,400 tok (1 alert)
30-day data points	134

Critical Observations (always visible)

• ⚠️ list_code_scanning_alerts bloat persists — a single alert costs ~1,400 tokens, ~80% of which is the full rule.help markdown (two multi-line Go examples) + full_description. The actionable signals (rule.id, security_severity_level=critical, location.path:line → pkg/workflow/awf_helpers.go:166) are buried. The help text should be fetched on-demand via get, not inlined per row.
• 🔒 get_me and actions_list unavailable — get_me returns 403 (token lacks /user scope), actions_list permission not granted in this run. Both are environmental restrictions, not tool-design regressions, and are scored 1 only because they could not be evaluated.
• ✅ List/discovery tools stay lean — search_repositories (185 tok), list_discussions (160 tok), and get_label (30 tok) carry no envelope bloat.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	8
Total Tokens (Today)	~3,075
Average Usefulness Rating	3.12 / 5 (3.83 excluding unavailable)
Best Rated Tool	search_repositories: 5/5
Worst Rated (design)	list_code_scanning_alerts / get_label: 3/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
search_repositories	search	⭐⭐⭐⭐⭐	Flat fields, all discovery signals, zero bloat
list_discussions	discussions	⭐⭐⭐⭐	Cursor pagination + totalCount, compact items
list_issues	issues	⭐⭐⭐⭐	Compact schema; size is body-driven
list_pull_requests	pull_requests	⭐⭐⭐⭐	Rich base/head+sha; redundant nested repo objects
get_label	labels	⭐⭐⭐	Tiny & clean but needs exact name first
list_code_scanning_alerts	code_security	⭐⭐⭐	Verbose rule.help bloat dominates payload
get_me	context	⭐	Unavailable (403) — normally 5
actions_list	actions	⭐	Unavailable (permission) — normally 4

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_label	object	1	color, description, id, name	Clean flat lookup
search_repositories	array	2	full_name, language, stars, forks, topics, open_issues	Best-in-class minimal_output
list_discussions	object	3	category, number, title, user.login, pageInfo, totalCount	GraphQL-style envelope
list_issues	array	3	number, title, state, labels, user.login, body	Labels as plain names
list_pull_requests	array	4	number, draft, merged, base/head ref+sha, labels	Nested repo objects redundant
list_code_scanning_alerts	array	4	rule.id, rule.help, security_severity_level, location	Help text bloat

Response Size Analysis

Toolset	Tokens (today)	Tool Tested
code_security	~1,400	list_code_scanning_alerts
pull_requests	~760	list_pull_requests
issues	~520	list_issues
search	~185	search_repositories
discussions	~160	list_discussions
labels	~30	get_label
context	0 (unavailable)	get_me
actions	0 (unavailable)	actions_list

30-Day Trend Summary

Metric	Value
Data Points	134 (2026-05-20 → 2026-06-09)
search_repositories rating	Stable at 5/5
list_code_scanning_alerts	Consistently 2–3/5, bloat unresolved
get_me availability	Persistently 403 across runs

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Recommendations

• Prefer for agents (rating 4–5): search_repositories, list_discussions, list_issues, list_pull_requests — actionable with lean schemas.
• Context-efficient picks (low tokens, high value): search_repositories (185 tok), list_discussions (160 tok), get_label (30 tok).
• Use sparingly / paginate hard: list_code_scanning_alerts — fetch full rule.help on demand rather than listing it inline; request a summary mode upstream.
• Tool-design ask: strip per-item static rule documentation from code-scanning list responses, and de-duplicate the nested repo object embedded in PR base/head for single-repo listings.

References: §27207220474

Generated by 🔍 GitHub MCP Structural Analysis · 141.3 AIC · ⌖ 17.8 AIC · ⊞ 12.5K · ◷

• expires on Jun 10, 2026, 4:57 AM UTC-08:00

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-06-10T12:57:14.196Z.

Closed by Workflow