[mcp-analysis] MCP Structural Analysis - 2026-04-16

[github-actions[bot]]

Analyzed 9 GitHub MCP tools across 9 toolsets on 2026-04-16. list_discussions and list_pull_requests remain the top performers (rating 5 and 4 respectively), while list_code_scanning_alerts and search_code continue to be the most context-expensive and least agentic-friendly tools. Two-day trending data now available.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	9
Total Tokens (Today)	~49,820
Average Usefulness Rating	3.0 / 5
Best Rated Tools	list_discussions, list_pull_requests (5 / 4)
Worst Rated Tools	get_me, list_code_scanning_alerts, search_code (1 / 2 / 2)
Historical Data Points	18 (2 days)

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
list_discussions	discussions	⭐⭐⭐⭐⭐	Compact, paginated, category-filterable — ideal for agents
list_pull_requests	pull_requests	⭐⭐⭐⭐	Excellent structure; full body can inflate on long PRs
list_issues	issues	⭐⭐⭐⭐	Good; issue body can balloon with debug/log content
get_file_contents	repos	⭐⭐⭐	Useful but no size guard — large files consume huge context
list_workflows	actions	⭐⭐⭐	per_page ignored; returns 30 by default with redundant URL fields
list_label	labels	⭐⭐⭐	Returns ~100 labels without pagination; GraphQL node IDs add bloat
list_code_scanning_alerts	code_security	⭐⭐	Exceeds 145k chars (~36k tokens); deeply nested; unusable for most agents
search_code	search	⭐⭐	Repository sub-object has 30+ URL templates per result; high per-item cost
get_me	context	⭐	403 forbidden in this integration context; not accessible

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	error	1	error	Integration token blocks user endpoint
list_issues	array	3	number, title, state, body, user, labels, created_at	user → login shallow; body unbounded
list_pull_requests	array	4	number, title, state, head, base, body, draft, html_url	head/base nest repo context
list_workflows	object	2	id, name, path, state, url, html_url, badge_url	Redundant url/html_url/badge_url per entry
list_discussions	object	3	number, title, category, user, created_at, html_url, pageInfo	Clean cursor pagination
list_code_scanning_alerts	array	5	number, rule, severity, location, most_recent_instance	location → path → region nested; deeply verbose
list_label	object	2	name, color, description, id	GraphQL node ID adds 50+ chars per label
search_code	object	4	name, path, sha, html_url, repository	repository sub-object embeds 30+ URL templates
get_file_contents	object	2	content, sha, mimeType, uri	Entire file returned inline; no truncation option

Response Size Analysis

Toolset	Avg Tokens (2-day)	Today
code_security	~36,000	36,000
repos	~4,850	5,500
actions	~2,625	3,000
labels	~2,250	2,250
search	~813	800
pull_requests	~405	500
issues	~723	625
discussions	~115	125
context	20	20

Tool-by-Tool Analysis (Today)

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	20	error	1	403 — integration token blocks /user
list_issues	issues	625	array/3	4	Issue body included in full
list_pull_requests	pull_requests	500	array/4	4	PR body included; nested head/base
list_workflows	actions	3,000	object/2	3	per_page=1 ignored; returns 30
list_discussions	discussions	125	object/3	5	Compact, cursor-paginated
list_code_scanning_alerts	code_security	36,000	array/5	2	145k chars, exceeds token limit
list_label	labels	2,250	object/2	3	~100 labels; no pagination
search_code	search	800	object/4	2	Repo sub-object has 30+ URL templates
get_file_contents	repos	5,500	object/2	3	README ~22k chars returned inline

30-Day Trend Summary

Metric	Value
Data Points	18 (2 days)
Avg Daily Total Tokens	~49,820
Trend Direction	Stable (consistent across 2 days)
Consistent Issues	code_security bloat, search_code repo sub-object, per_page ignored in actions
Consistent Strengths	discussions compact structure, PR/issue data actionable

Recommendations

High-value tools (rating 4–5) — prefer these:

• list_discussions — best token efficiency with full pagination
• list_pull_requests — excellent nested context for PR workflows
• list_issues — good for issue triage; constrain per_page

Tools needing improvement:

• list_code_scanning_alerts — needs per_page/pagination support; current size makes it unusable
• search_code — repository sub-object should be reduced to full_name + html_url
• list_workflows — per_page parameter should be respected
• list_label — needs pagination; GraphQL node IDs could be omitted

Context-efficient tools (low tokens, high rating):

• list_discussions (125 tokens, rating 5)

Context-heavy tools to use with caution:

• list_code_scanning_alerts (36,000 tokens) — avoid unless strictly needed
• get_file_contents (5,500+ tokens) — only fetch small known files

Visualizations

Response Size by Toolset

Usefulness Ratings by Tool

Daily Token Usage Trend

Token Size vs Usefulness

References:

• §24509507969

Generated by GitHub MCP Structural Analysis · ● 350.1K · ◷

• expires on Apr 17, 2026, 12:30 PM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #26864.