[mcp-analysis] MCP Structural Analysis - 2026-05-06

[github-actions[bot]]

Today's analysis examined 9 GitHub MCP tools across 9 toolsets, measuring response sizes and rating their usefulness for agentic workflows. Total token cost: 14,200 tokens for one call per tool. Average usefulness rating: 3.3/5. Standout finding: list_code_scanning_alerts is severely bloated (6,000 tokens for 4 alerts) due to repeating full CWE documentation in every response.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	9
Toolsets Covered	context, repos, issues, pull_requests, actions, code_security, discussions, labels, search
Total Tokens (Today)	14,200
Average Usefulness Rating	3.3/5
Best Rated Tool	list_discussions: 5/5
Most Bloated Tool	list_code_scanning_alerts: 6,000 tokens / rating 2/5
Auth Failure	get_me returned 403 with integration token

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
list_discussions	discussions	⭐⭐⭐⭐⭐	Excellent — minimal payload, cursor pagination, totalCount, category filter
list_issues	issues	⭐⭐⭐⭐	Good — complete issue data; body can bloat but bounded by perPage
list_pull_requests	pull_requests	⭐⭐⭐⭐	Good — head/base refs and draft/merged flags immediately actionable
get_label	labels	⭐⭐⭐⭐	Good — very compact single-record response with all essential fields
search_issues	search	⭐⭐⭐⭐	Good — rich results with reactions, comment counts, full label metadata
get_file_contents	repos	⭐⭐⭐	Adequate — no size limit; massive for large files; SHA useful for caching
list_workflows	actions	⭐⭐⭐	Adequate — per_page param ignored, returns 29 workflows instead of 1; no trigger info
list_code_scanning_alerts	code_security	⭐⭐	Limited — rule.help field embeds full CWE docs (~1,500 tokens) repeated per identical alert
get_me	context	⭐	Poor — 403 error with integration token; not accessible in agentic workflow context

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	error	1	error_message	Auth failure
get_file_contents	object	2	content, sha, mimeType, uri	Entire file in one response
list_issues	array	3	number, title, body, state, labels, user	Labels embed full metadata
list_pull_requests	array	3	number, title, body, head, base, merged, draft	head/base are nested objects
list_workflows	object	2	id, name, path, state, created_at	Paginator ignored in test
list_code_scanning_alerts	array	3	number, rule, tool, state, most_recent_instance	rule.help is extreme bloat
list_discussions	object	2	number, title, category, html_url, totalCount	Cleanest schema observed
get_label	object	1	name, color, description, id	Minimal depth, ideal
search_issues	array	3	number, title, body, state, labels, reactions	Full label objects in results

Response Size Analysis

Toolset	Tool	Tokens	Rank
code_security	list_code_scanning_alerts	6,000	Largest
repos	get_file_contents	3,750	2nd
actions	list_workflows	2,250	3rd
pull_requests	list_pull_requests	800	4th
issues	list_issues	600	5th
search	search_issues	500	6th
discussions	list_discussions	200	7th
labels	get_label	50	Smallest
context	get_me	50	Smallest (error)

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	50	error/object	1	403 with integration token
get_file_contents	repos	3,750	object depth=2	3	Full README, no truncation
list_issues	issues	600	array depth=3	4	perPage respected, good data
list_pull_requests	pull_requests	800	array depth=3	4	draft/merged flags useful
list_workflows	actions	2,250	object depth=2	3	per_page param not applied
list_code_scanning_alerts	code_security	6,000	array depth=3	2	rule.help bloat per alert
list_discussions	discussions	200	object depth=2	5	Best-in-class efficiency
get_label	labels	50	object depth=1	4	Tiny and complete
search_issues	search	500	array depth=3	4	Rich with reactions/comments

30-Day Trend Summary

Metric	Value
Data Points	9 (day 1 of tracking)
Total Tokens (2026-05-06)	14,200
Average Rating	3.3/5
Trend	Baseline established — no prior data

Recommendations

High-value tools (rating 4–5) — prefer for agentic workflows:

• list_discussions, list_issues, list_pull_requests, get_label, search_issues

Tools needing improvement:

• list_code_scanning_alerts — strip rule.help from the default response; surface it only on get_code_scanning_alert. This alone would reduce alert list responses by ~80%.
• list_workflows — honour per_page parameter to prevent context overflow when a repo has hundreds of workflows.
• get_me — expose a minimal profile endpoint accessible with the standard Actions integration token.

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

• get_label (50 tokens, 4/5), list_discussions (200 tokens, 5/5)

Context-heavy tools — use with caution or cache results:

• list_code_scanning_alerts (6,000 tokens), get_file_contents (3,750 tokens for large files)

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References:

• §25435035923

Generated by GitHub MCP Structural Analysis · ● 471.8K · ◷

• expires on May 7, 2026, 12:38 PM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #30824.