[mcp-analysis] GitHub MCP Structural Analysis — 2026-06-04

[github-actions[bot]]

Today's sampling tested 9 GitHub MCP tools across 9 toolsets (1 representative tool each, minimal params). Eight returned data; get_me (context) was again blocked with a 403 "Resource not accessible by integration" and actions_list (actions) was blocked by an ungranted permission this run — both environmental limits, not tool-design regressions.

Key findings: get_file_contents and search_repositories remain the standouts (5/5 — complete, immediately actionable, context-efficient). The biggest context cost continues to be per-item envelope bloat: a single list_code_scanning_alerts result spends ~1,100 tokens because it embeds the full CodeQL rule help text + two Go code examples for one alert. Average usefulness today is 3.33/5; the 11-day trend is stable (slope ≈ 0).

Snapshot

Metric	Value
Tools analyzed today	9 (8 evaluated + 1 unavailable)
Total tokens (today, sampled)	~4,300
Average usefulness	3.33 / 5
Best rated	get_file_contents, search_repositories — 5/5
Lowest rated	get_me, actions_list — 1/5 (both unavailable this run)
Trend window	108 data points over 11 days
Rating trend	Stable

Recommendations

• Reach for first (high value, context-efficient): search_repositories (minimal_output default), get_file_contents (returns SHA for follow-up edits), list_discussions (lean list view + totalCount + cursor pagination).
• Use with care (high token cost): list_code_scanning_alerts — request specific alert numbers or a summary view rather than listing; the embedded rule help dominates cost. list_pull_requests and list_issues are content-driven — the body inflates cost, so prefer narrow perPage and field-light queries for triage.
• Discovery-only: get_label is tiny but needs the exact name first — prefer list_label to discover.
• Environment note: get_me and actions_list were not testable in this run. No action required; tracked as unavailable so trend averages aren't skewed (excluded from token charts).

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	9
Total Tokens (Today, sampled)	~4,300
Average Usefulness Rating	3.33 / 5
Best Rated Tool	get_file_contents: 5/5
Worst Rated Tool	get_me: 1/5 (unavailable — 403)

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
get_file_contents	repos	⭐⭐⭐⭐⭐	Full text + SHA + uri in one call; immediately actionable for reads/edits
search_repositories	search	⭐⭐⭐⭐⭐	minimal_output flat fields, all discovery signals, no URL bloat
list_issues	issues	⭐⭐⭐⭐	Compact schema (labels as names, user.login); body drives cost
list_pull_requests	pull_requests	⭐⭐⭐⭐	Rich PR-ops data (base/head ref+sha, draft/merged, assignees); body-heavy
list_discussions	discussions	⭐⭐⭐⭐	Lean items + cursor pagination + totalCount; minor JSON-string envelope
list_code_scanning_alerts	code_security	⭐⭐⭐	Actionable rule.id/severity/location, but full help text is redundant bloat
get_label	labels	⭐⭐⭐	Minimal complete object; low standalone value (needs exact name)
get_me	context	⭐	403 not accessible by integration — unavailable this run (would normally be 5)
actions_list	actions	⭐	Permission not granted this run — could not evaluate

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_file_contents	object	2	content, text, sha, uri, mimeType	Resource envelope; SHA enables follow-up writes
search_repositories	array	2	full_name, language, stargazers_count, topics, open_issues_count	Flat, discovery-optimized
list_issues	array	3	number, title, state, labels, user, timestamps, body	Labels as plain names; user.login only
list_pull_requests	array	4	number, state, draft, merged, base, head, user, assignees	Depth 4 from base/head.repo objects
list_discussions	object	3	discussions[], category, pageInfo, totalCount	Cursor pagination; content/text wrapper
list_code_scanning_alerts	array	4	number, rule, tool, state, most_recent_instance, location	Embeds full rule help + code examples
get_label	object	1	name, color, description, id	Flat, tiny
get_me	error	1	error	403 — unavailable
actions_list	unavailable	0	—	Permission not granted

Response Size Analysis (this run)

Tool	Toolset	Tokens
list_pull_requests	pull_requests	~1,600
list_code_scanning_alerts	code_security	~1,100
get_file_contents	repos	~665
list_issues	issues	~600
list_discussions	discussions	~155
search_repositories	search	~123
get_label	labels	~33
get_me	context	~24 (error)
actions_list	actions	n/a (unavailable)

Average Tokens by Toolset (30-day history, evaluated calls)

Toolset	Avg Tokens	Samples
actions	3,617	3
issues	2,813	11
repos	1,995	16
code_security	1,490	11
search	1,178	17
labels	1,096	12
pull_requests	896	11
discussions	155	11
context	28	11

(History averages include heavier tool variants sampled on prior days — e.g. list_label vs get_label, full-output search — so they run higher than today's specific picks.)

30-Day Trend Summary

Metric	Value
Data Points	108
Days Covered	11 (2026-05-20 → 2026-06-04)
Average Daily Tokens (sampled)	~12,287
Average Rating Trend	Stable (≈ 3.3–3.75/day, slope ≈ 0)

Recommendations (detail)

• High-value tools (4–5): get_file_contents, search_repositories, list_issues, list_pull_requests, list_discussions.
• Tools needing improvement: list_code_scanning_alerts — the verbose per-alert help text should be opt-in for list/triage views.
• Context-efficient (low tokens, high rating): search_repositories (~123 tok, 5/5), list_discussions (~155 tok, 4/5).
• Context-heavy (high tokens): list_pull_requests and list_code_scanning_alerts — body / embedded help dominate cost.

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References: §26952805714

Generated by 🔍 GitHub MCP Structural Analysis · opus48 8.3M · ◷

• expires on Jun 5, 2026, 1:00 PM UTC

[pelikhan]

/q optimize this agentic workflow

[github-actions[bot]]

🔧 Pay attention, 007! Q is preparing your gadgets for this discussion comment...