[mcp-analysis] MCP Structural Analysis - 2026-06-02

[github-actions[bot]]

Overview

Analyzed 12 GitHub MCP tool responses across 9 toolsets today, totaling ~13,199 tokens with an average usefulness of 3.75/5 — the highest daily average in the current 30-day window (range 3.2–3.75 over 91 data points).

Key findings:

• Best-in-class (5/5): list_discussions (180 tok), search_repositories (135 tok), get_label (36 tok), list_branches (28 tok) — compact, flat, all signal, no bloat.
• Worst for agents (2/5): search_issues (~4,125 tok) and list_code_scanning_alerts (~1,190 tok). Both are dominated by static/envelope bloat rather than useful signal.
• Standout anti-pattern: for the same single issue, search_issues (full REST envelope: reactions, ~10 *_url fields, 15-field user, expanded label objects) costs ~20% more than list_issues and far more than needed. Prefer list_issues for issue listing.
• Caveats this run: get_me returned 403 Resource not accessible by integration (environmental, not a quality regression); actions_list permission was not granted, so the actions toolset was not evaluated today.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	12 (1 errored: get_me)
Total Tokens (Today)	13,199
Average Usefulness Rating	3.75 / 5
Best Rated Tool	list_discussions / search_repositories / get_label / list_branches: 5/5
Worst Rated Tool	search_issues / list_code_scanning_alerts: 2/5 (get_me errored)
Most Context-Efficient	list_branches: 28 tokens @ 5/5
Heaviest Response	search_issues: ~4,125 tokens @ 2/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
list_discussions	discussions	⭐⭐⭐⭐⭐	GraphQL envelope + cursor pagination + totalCount, minimal per-item fields
search_repositories	search	⭐⭐⭐⭐⭐	minimal_output flat fields; all discovery signals, no URL bloat
get_label	labels	⭐⭐⭐⭐⭐	Tiny single-purpose lookup; color + description sufficient
list_branches	repos	⭐⭐⭐⭐⭐	name + sha + protected; exactly what branch ops need
get_file_contents	repos	⭐⭐⭐⭐	File text in resource envelope w/ mimeType + uri + SHA (great for edits)
list_commits	repos	⭐⭐⭐⭐	sha + message + author; minor avatar/profile URL bloat
list_label	labels	⭐⭐⭐⭐	Full lean label set + totalCount; ignores perPage (fixed cost)
list_issues	issues	⭐⭐⭐⭐	Compact, actionable; tokens driven by issue body, not schema
list_pull_requests	pull_requests	⭐⭐⭐⭐	base/head ref+sha, draft, merged, labels — strong for PR ops
list_code_scanning_alerts	code_security	⭐⭐	~80% of payload is static rule.help markdown; verbose for triage
search_issues	search	⭐⭐	Full REST envelope bloat (reactions, *_url, full user) + large body
get_me	context	⭐	Errored (403) in sandbox — could not evaluate (environmental)

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_label	object	1	id, name, color, description	Flattest, cleanest
list_branches	array	1	name, sha, protected	Minimal, ideal
list_discussions	object	3	discussions, pageInfo, totalCount, number, title, category, user	Cursor pagination envelope
search_repositories	array	2	full_name, language, stargazers_count, topics, open_issues_count	Flat discovery signals
get_file_contents	object	3	content, resource, mimeType, text, uri, sha	SHA enables follow-up edits
list_commits	array	3	sha, commit.message, author, committer	Duplicated author/committer objects
list_label	object	2	labels, totalCount, name, color, description, id	Lean per-item
list_issues	array	3	number, title, state, labels, user, timestamps, body	Labels as plain names
list_pull_requests	array	4	number, state, draft, merged, base, head, labels, user	base/head.repo nesting
list_code_scanning_alerts	array	5	number, rule, tool, state, most_recent_instance, location	Deep; rule.help dominates
search_issues	array	4	number, title, labels, body, user, reactions, *_url	REST envelope bloat

Response Size Analysis (today, avg tokens per toolset)

Toolset	Avg Tokens	Tools Tested
issues	3,450	list_issues
search	2,130	search_repositories, search_issues
pull_requests	1,550	list_pull_requests
code_security	1,190	list_code_scanning_alerts
labels	843	get_label, list_label
repos	288	get_file_contents, list_branches, list_commits
discussions	180	list_discussions
context	20 (errored)	get_me

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
list_branches	repos	28	array/1	5	Maximally compact, zero bloat
get_label	labels	36	object/1	5	Single-purpose lookup
search_repositories	search	135	array/2	5	All discovery signals, no URLs
list_commits	repos	160	array/3	4	Minor avatar/profile-URL bloat
list_discussions	discussions	180	object/3	5	Cursor pagination + totalCount
get_file_contents	repos	675	object/3	4	Resource envelope + SHA
list_code_scanning_alerts	code_security	1,190	array/5	2	rule.help static docs dominate
list_pull_requests	pull_requests	1,550	array/4	4	Body inflates; rich PR-ops fields
list_label	labels	1,650	object/2	4	Full set; perPage ignored
list_issues	issues	3,450	array/3	4	Body-driven cost; clean schema
search_issues	search	4,125	array/4	2	REST envelope + body bloat
get_me	context	20	error	1	403 — environmental, not evaluated

30-Day Trend Summary

Metric	Value
Data Points	91 (9 days, 2026-05-20 → 2026-06-02)
Average Daily Tokens / Tool	~1,366
Today's Avg Tokens / Tool	~1,100 (mid-range)
Average Rating Trend	Stable (3.2–3.75); today 3.75 = window high

Recommendations

• High-value tools (rating 4–5): list_discussions, search_repositories, get_label, list_branches, get_file_contents, list_commits, list_label, list_issues, list_pull_requests. Reach for these freely.
• Tools needing improvement:
  • list_code_scanning_alerts — add a summary mode that strips rule.help/full_description (static rule docs repeat across alerts sharing a rule). Would cut the bulk of its tokens.
  • search_issues — drop the REST envelope (reactions, *_url, full user, node_id) for agent use; it duplicates list_issues at higher cost.
• Context-efficient (low tokens, high rating): list_branches (28), get_label (36), search_repositories (135), list_discussions (180).
• Context-heavy (use deliberately): search_issues (~4,125), list_issues (~3,450) — cost is mostly issue body; request narrow result sets.
• Selection rule: prefer list_issues over search_issues whenever you don't need full-text search — same data, materially fewer tokens.

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References:

• §26821399897

Generated by 🔍 GitHub MCP Structural Analysis · opus48 1.8M · ◷

• expires on Jun 3, 2026, 1:10 PM UTC

[github-actions[bot]]

This discussion has been marked as outdated by GitHub MCP Structural Analysis.

A newer discussion is available at Discussion #36645.