[mcp-analysis] MCP Structural Analysis - 2026-06-24

[github-actions[bot]]

GitHub MCP Structural Analysis — 2026-06-24

Quantitative (response size) + qualitative (schema & agentic usefulness) analysis of 9 representative GitHub MCP tools, one per available toolset.

Key findings today:

• Best for agents: search_repositories ⭐⭐⭐⭐⭐ — flat, ~200 tokens, every discovery signal, zero bloat.
• Worst: get_me ⭐ — returns 403 Resource not accessible by integration; the GITHUB_TOKEN has no user scope, so identity lookups are unavailable in this auth context.
• Two pagination/bloat regressions worth flagging:
  • list_workflows ignored per_page=1 and returned 30 entries (~1650 tokens, total_count=324). Each carries badge_url/node_id/api-url that agents rarely use.
  • list_issues honored perPage=1 but inlined the issue's full multi-KB body (a token-audit report) → ~1400 tokens for one issue. A label-parsing artifact also split a single audit,token-usage label into two tokens.
• list_code_scanning_alerts still inlines the full rule.help CWE markdown per result (~560 tokens for ONE alert #635 js/insecure-temporary-file, high severity, open since 06-22).

Snapshot: 9 tools · 5,550 tokens total · avg usefulness 3.33/5. 30-day window holds 191 data points across 21 days; rating trend stable (3.40 → 3.33).

Recommendations (tool selection for agentic work)

• High-value (4–5): search_repositories, list_pull_requests, list_discussions, get_label, get_file_contents — prefer these for discovery/triage.
• Context-efficient sweet spot (low tokens, high rating): get_label (45t/r4), list_discussions (210t/r4), search_repositories (200t/r5).
• Context-heavy — page/scope tightly: list_workflows (1650t, ignores per_page), list_issues (1400t, body inlined).
• Needs improvement: list_code_scanning_alerts (move rule.help to on-demand lookup), get_me (auth-limited — derive identity elsewhere).

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	9
Total Tokens (Today)	5,550
Average Usefulness Rating	3.33/5
Best Rated Tool	search_repositories: 5/5
Worst Rated Tool	get_me: 1/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
search_repositories	search	⭐⭐⭐⭐⭐	Flat, complete discovery signals; best context efficiency
list_pull_requests	pull_requests	⭐⭐⭐⭐	Rich actionable flags (draft/merged/refs); nested repo dup
list_discussions	discussions	⭐⭐⭐⭐	Cursor pagination + totalCount; triage-only, needs follow-up
get_label	labels	⭐⭐⭐⭐	Flat & complete; narrow scope caps leverage
get_file_contents	repos	⭐⭐⭐⭐	mimeType + text + (redacted) uri; cost scales with file
list_workflows	actions	⭐⭐⭐	Ignores per_page; badge_url/node_id bloat
list_issues	issues	⭐⭐⭐	Full body inlined; label split artifact
list_code_scanning_alerts	code_security	⭐⭐	Full rule.help markdown inlined per result
get_me	context	⭐	403 — not accessible by this integration

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	object	1	error.text, isError	Error envelope, not a profile
list_workflows	object	3	total_count, id, name, path, state, html_url	Clean but pagination ignored
list_issues	array	3	number, title, state, labels, user.login, body	Body bloat; label split
list_pull_requests	array	4	number, state, draft, merged, head.ref, base.ref	base.repo + head.repo duplicated
list_code_scanning_alerts	array	5	number, rule.id, security_severity_level, location.path	rule.help inlined
list_discussions	object	3	number, title, category.name, pageInfo, totalCount	GraphQL-style envelope
get_label	object	1	name, color, description, id	Flat, proportionate
get_file_contents	array	3	resource.text, resource.mimeType, resource.uri	SHA-confirm + resource obj
search_repositories	array	2	full_name, language, stargazers_count, topics, open_issues_count	Flat, no permissions bloat

Response Size Analysis (today, 1 tool per toolset)

Toolset	Tokens	Tool
actions	1650	list_workflows
issues	1400	list_issues
pull_requests	760	list_pull_requests
repos	700	get_file_contents
code_security	560	list_code_scanning_alerts
discussions	210	list_discussions
search	200	search_repositories
labels	45	get_label
context	25	get_me (error)

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
search_repositories	search	200	array/2	5	minimal_output flat fields; stars 4689 (+2), open_issues 264 (−26)
list_pull_requests	pull_requests	760	array/4	4	head/base refs+sha actionable; nested repo desc duplicated
list_discussions	discussions	210	object/3	4	totalCount 7664 (+43/day); proper cursor pagination
get_label	labels	45	object/1	4	Complete single-label payload
get_file_contents	repos	700	array/3	4	.gitignore; resource carries uri + mimeType
list_workflows	actions	1650	object/3	3	per_page=1 ignored → 30 rows; total_count 324
list_issues	issues	1400	array/3	3	Full body inlined; label split into 2 tokens
list_code_scanning_alerts	code_security	560	array/5	2	rule.help CWE-377 markdown inlined; alert #635 high/open
get_me	context	25	object/1	1	403 not accessible by integration

30-Day Trend Summary

Metric	Value
Data Points	191
Distinct Days	21
Average Daily Tokens	8,541
Average Rating Trend	Stable (3.40 → 3.33)

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References: §28098801795

Generated by 🔍 GitHub MCP Structural Analysis · 236.7 AIC · ⌖ 29 AIC · ⊞ 11.7K · ◷

• expires on Jun 25, 2026, 4:51 AM UTC-08:00