[mcp-analysis] MCP Structural Analysis - 2026-06-16

[github-actions[bot]]

GitHub MCP Structural Analysis — 2026-06-16

Tested 9 representative tools across the available GitHub MCP toolsets, measuring response size (tokens), schema structure, and usefulness for agentic work (1–5). Today's run consumed ≈8,400 tokens at an average usefulness of 3.33/5.

Headlines today:

• 🥇 Best (5/5): search_repositories and get_file_contents — flat, signal-dense, immediately actionable schemas.
• 🥉 Worst (1/5): get_me (403 "Resource not accessible by integration" — integration token lacks /user scope) and actions_list (permission denied — actions toolset not in this workflow's allow-list).
• 📈 Notable size shift: get_file_contents jumped to ~3,400 tokens (vs 340–700 recently) — not tool bloat, but because README.md itself ballooned with an auto-generated ~140-person "Community Contributions" block. Cost is 100% file-dependent.
• 🔁 Persistent structural bloat: list_code_scanning_alerts still inlines the full educational rule.help CWE markdown per alert (~70% of its payload). Tolerable at 1 alert (~1,300t), prohibitive at scale.

Key Metrics

Metric	Value
Tools Analyzed (today)	9
Total Tokens (today)	≈8,400
Average Usefulness (today)	3.33 / 5
Best Rated	search_repositories, get_file_contents — 5/5
Worst Rated	get_me, actions_list — 1/5
30-day Data Points	178 across 19 days
Avg Daily Tokens (30d)	≈10,032
Rating Trend (30d)	3.6 → 3.33 (≈stable, slight dip)

Recommendations

• High-value, context-efficient (use freely): search_repositories (~160t, 5/5), list_discussions (~180t, 4/5), get_label (~40t, 3/5) — minimal payloads with proper pagination envelopes.
• High-value but heavy (page/scope carefully): list_issues (~2,300t) and list_pull_requests (~1,000t) carry full bodies + redundant nested repo objects — request narrower fields or fewer items when only triaging.
• Needs upstream improvement: list_code_scanning_alerts should move rule.help/full_description to an on-demand lookup rather than inlining per result.
• File-dependent, not tool-dependent: get_file_contents cost tracks the file — prefer ranged/path-targeted reads on large docs like README.md.
• Unavailable in this context: get_me (403) and actions_list (not granted) — agents should fall back to the injected github-context for identity rather than get_me.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	9
Total Tokens (Today)	≈8,400
Average Usefulness Rating	3.33/5
Best Rated Tool	get_file_contents / search_repositories: 5/5
Worst Rated Tool	get_me / actions_list: 1/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
get_file_contents	repos	⭐⭐⭐⭐⭐	Clean envelope (sha-pinned uri + raw text); cost is file-driven
search_repositories	search	⭐⭐⭐⭐⭐	Flat, every discovery signal present, zero bloat
list_issues	issues	⭐⭐⭐⭐	Actionable triage fields; full body inflates cost
list_pull_requests	pull_requests	⭐⭐⭐⭐	Rich PR-ops payload; redundant nested repo objects
list_discussions	discussions	⭐⭐⭐⭐	Cursor pagination + totalCount; no body (needs follow-up)
list_code_scanning_alerts	code_security	⭐⭐⭐	Excellent finding fields, but inlined help markdown bloat
get_label	labels	⭐⭐⭐	Complete but low standalone value (validation primitive)
get_me	context	⭐	FAILED 403 — integration token lacks user scope
actions_list	actions	⭐	DENIED — toolset not in workflow allow-list

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	object	1	error	Error envelope only
get_file_contents	object	2	mimeType, text, uri(sha), sha	Clean; cost = file size
list_issues	array	3	number, title, state, labels, user.login, body	Body-heavy
list_pull_requests	array	4	number, base/head ref+sha, draft, merged, labels	Nested repo redundancy
actions_list	none	0	—	Permission denied
list_code_scanning_alerts	array	5	rule.id, security_severity_level, location.path+line, message.text	Help markdown inlined
list_discussions	object	3	number, title, category.name, pageInfo, totalCount	GraphQL-style envelope
get_label	object	1	name, color, description, id	Flat 4-field
search_repositories	array	2	full_name, stars, forks, language, topics, open_issues	Best-in-class

Response Size Analysis (30-day avg by toolset)

Toolset	Avg Tokens	Representative Tool
issues	≈2,392	list_issues
repos	≈1,929	get_file_contents
code_security	≈1,797	list_code_scanning_alerts
pull_requests	≈923	list_pull_requests
search	≈854	search_repositories / search_issues
actions	≈723	actions_list / list_workflows (mostly denied → low)
labels	≈671	get_label / list_label
discussions	≈162	list_discussions
context	≈24	get_me (failing)

Tool-by-Tool Analysis (today)

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	20	object	1	403 not accessible by integration; rely on github-context
get_file_contents	repos	3400	object	5	README ballooned w/ contributor list; schema still clean
list_issues	issues	2300	array	4	Full markdown body dominates; triage fields clean
list_pull_requests	pull_requests	1000	array	4	base/head ref+sha actionable; nested repo redundancy
actions_list	actions	0	none	1	Permission denied — toolset not granted
list_code_scanning_alerts	code_security	1300	array	3	1 critical alert; ~70% payload is inlined rule.help
list_discussions	discussions	180	object	4	totalCount=7331 (+39 vs 06-15); cursor pagination
get_label	labels	40	object	3	Flat name/color/description/id
search_repositories	search	160	array	5	stars 4629 (+5 vs 06-15); flat discovery signals

30-Day Trend Summary

Metric	Value
Data Points	178
Days Covered	19 (2026-05-20 → 2026-06-16)
Average Daily Tokens	≈10,032
Average Rating Trend	Stable (3.6 → 3.33), slight recent dip driven by code_security help-bloat and the get_me/actions denials

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References: §27620807014

Generated by 🔍 GitHub MCP Structural Analysis · 212.5 AIC · ⌖ 34.9 AIC · ⊞ 11.7K · ◷

• expires on Jun 17, 2026, 5:41 AM UTC-08:00