[mcp-analysis] MCP Structural Analysis - 2026-06-17

[github-actions[bot]]

GitHub MCP Structural Analysis — 2026-06-17

Analyzed 8 GitHub MCP tools (one per available toolset) for response size, schema structure, and usefulness for agentic work. Average usefulness this run is 3.88/5 across ~4,990 tokens. Trend dataset now holds 186 data points over 20 days.

Key findings

• Top-rated (5/5): get_file_contents, get_label, search_repositories — clean, flat, immediately actionable, size proportional to the request.
• Lowest (1/5): get_me — returns 403 Resource not accessible by integration; the GitHub App/Actions token cannot resolve the authenticated user, so identity-dependent flows must be skipped.
• Biggest structural waste: list_code_scanning_alerts (1,350 tok) still inlines the full rule.help CWE markdown (~70% of payload) on every result. Fine at 1 alert; prohibitive at scale.
• Recurring bloat: list_pull_requests embeds a redundant nested repo object in both base and head within a single-repo listing; list endpoints (list_issues, list_pull_requests) ship full bodies, which dominate token cost when paging.
• Best context efficiency: get_label (45 tok), search_repositories (175 tok), list_discussions (190 tok, with proper cursor pagination).

Note: the actions toolset (list_workflows) was attempted but blocked by integration permissions this run, so it is excluded from today's dataset (not fabricated). The Usefulness by Toolset chart reflects the corrected 8-tool set; the other three charts include a negligible, near-hidden placeholder point for it.

Recommendations

• Use freely (rating 4–5): get_file_contents, get_label, search_repositories, list_discussions, list_issues, list_pull_requests.
• Needs improvement: list_code_scanning_alerts — move rule.help to an on-demand lookup instead of per-result inlining. get_me — auth-scope mismatch for App/Actions tokens.
• Most context-efficient (low tokens, high rating): get_label (45/5), search_repositories (175/5), list_discussions (190/4).
• Context-heavy: list_code_scanning_alerts (1,350, bloat), get_file_contents (1,300, intrinsic to file), list_pull_requests (1,100).

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	8
Total Tokens (Today)	~4,990
Average Usefulness Rating	3.88 / 5
Best Rated Tool	get_file_contents / get_label / search_repositories: 5/5
Worst Rated Tool	get_me: 1/5 (403 auth)

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
get_file_contents	repos	⭐⭐⭐⭐⭐	Clean MCP resource envelope, content-addressed URI + SHA; size scales only with the file
get_label	labels	⭐⭐⭐⭐⭐	Flat object, zero nesting, zero bloat
search_repositories	search	⭐⭐⭐⭐⭐	minimal_output carries every discovery signal, no permissions bloat
list_issues	issues	⭐⭐⭐⭐	Triage essentials clean; full body dominates cost
list_pull_requests	pull_requests	⭐⭐⭐⭐	Rich PR-ops payload; redundant nested repo objects in base/head
list_discussions	discussions	⭐⭐⭐⭐	Cursor pagination + totalCount; no body/labels (needs follow-up)
list_code_scanning_alerts	code_security	⭐⭐⭐	Excellent finding fields; ~70% payload is inlined rule.help
get_me	context	⭐	403 not accessible by integration

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	object	1	error	isError envelope only (403)
get_file_contents	object	3	resource.text, mimeType, uri, sha	Content-addressed resource
list_issues	array	3	number, title, state, labels, user.login, body	Body inflates size
list_pull_requests	array	4	number, state, draft, merged, base/head ref+sha, labels, assignees	Nested repo dup
list_code_scanning_alerts	array	5	rule.id, security_severity_level, location.path+start_line, message.text, html_url	Deeply nested; help bloat
list_discussions	object	3	discussions[].number, title, category.name, user.login, pageInfo, totalCount	GraphQL-style envelope
get_label	object	1	name, color, description, id	Minimal, clean
search_repositories	array	2	full_name, language, stargazers_count, forks_count, topics, open_issues_count	Flat discovery signals

Response Size Analysis

Toolset	Avg Tokens (today)	Tool Tested
code_security	1,350	list_code_scanning_alerts
repos	1,300	get_file_contents
pull_requests	1,100	list_pull_requests
issues	800	list_issues
discussions	190	list_discussions
search	175	search_repositories
labels	45	get_label
context	30	get_me (error)

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	30	object/1	1	403 Resource not accessible by integration; non-functional in this read-only App context
get_file_contents	repos	1,300	object/3	5	go.mod; size intrinsic to file, content-addressed uri + SHA, ready for read/patch
list_issues	issues	800	array/3	4	#39787; lighter body than 06-16; body still dominates cost
list_pull_requests	pull_requests	1,100	array/4	4	#39784; base/head ref+sha actionable; redundant nested repo objects
list_code_scanning_alerts	code_security	1,350	array/5	3	#600 go/unsafe-quoting (unchanged since 05-20); ~70% is rule.help markdown
list_discussions	discussions	190	object/3	4	totalCount 7,375 (+44 vs 06-16); cursor pagination; no body/labels
get_label	labels	45	object/1	5	'bug'; flat name/color/description/id
search_repositories	search	175	array/2	5	stars 4,633 (+4 vs 06-16); minimal_output best-in-class

30-Day Trend Summary

Metric	Value
Data Points	186
Days Covered	20 (2026-05-20 → 2026-06-17)
Average Daily Tokens	~9,781
Average Rating Trend	Stable (first-5-days 3.53 → last-5-days 3.41)

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References: §27690853261

Generated by 🔍 GitHub MCP Structural Analysis · ◷

• expires on Jun 18, 2026, 5:13 AM UTC-08:00