[mcp-analysis] MCP Structural Analysis - 2026-06-01

[github-actions[bot]]

Daily structural analysis of GitHub MCP tool responses — measuring response size (tokens), schema shape, and usefulness for autonomous agents. Run §26758897645 on github/gh-aw.

Overview

8 representative tools were probed with minimal parameters. Average usefulness this run: 3.5/5. The split is sharp and consistent with the 30-day history: lightweight lookup/browse tools (list_discussions, get_label, search_repositories) are best-in-class (5/5, all under ~125 tokens), while list_code_scanning_alerts remains the heaviest payload (~2,375 tokens for just 2 alerts) because each alert embeds full rule.help markdown.

Two context-specific findings stood out this run:

• get_me (context) returned 403 Resource not accessible by integration — unusable under the workflow's integration token. Agents on integration tokens should derive identity from the workflow context rather than calling it.
• list_issues was withheld by integrity policy (issue create_pull_request (cross-repo): PR is opened, then a second validation fails ERR_VALIDATION 'not in the allowed-repos list', marking the run failed #36242 below the approved threshold), returning an empty array plus a filter notice. An empty result here does not mean "no issues" — agents must account for silently filtered items.
• list_workflows (actions) could not be tested — permission was not granted in this run, so the actions toolset is absent from today's metrics.

Key Metrics

Metric	Value
Tools analyzed	8
Total tokens (today)	3,835
Average usefulness	3.5 / 5
Best rated	list_discussions, get_label, search_repositories — 5/5
Worst rated	get_me — 1/5 (403 error)
30-day data points	79 across 8 days

Recommendations

• High-value, context-efficient (use freely): list_discussions, get_label, search_repositories — all 5/5 at ≤125 tokens. Ideal first-choice tools for browsing and discovery.
• Strong but heavier: list_pull_requests (4/5, ~460 tok) and get_file_contents (4/5, scales with file size) are richly actionable; the embedded SHA in file contents is valuable for follow-up edits.
• Use with care / prefer narrower calls: list_code_scanning_alerts (2/5) — request a single alert or a summary view; ~80% of its payload is static rule.help docs repeated across alerts. list_issues (2/5 this run) — treat empty results skeptically due to integrity filtering.
• Avoid on integration tokens: get_me — derive identity from workflow context instead.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	8
Total Tokens (Today)	3,835
Average Usefulness Rating	3.5/5
Best Rated Tool	list_discussions: 5/5
Worst Rated Tool	get_me: 1/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
list_discussions	discussions	⭐⭐⭐⭐⭐	Cursor-paginated envelope, minimal fields — excellent
get_label	labels	⭐⭐⭐⭐⭐	Tiny single-purpose lookup
search_repositories	search	⭐⭐⭐⭐⭐	minimal_output flat fields, all discovery signals
list_pull_requests	pull_requests	⭐⭐⭐⭐	Rich and actionable; base/head repo objects duplicated
get_file_contents	repos	⭐⭐⭐⭐	Full file + SHA/uri envelope; scales with file size
list_code_scanning_alerts	code_security	⭐⭐	rule.help markdown bloat; ~80% static repeated docs
list_issues	issues	⭐⭐	Withheld by integrity policy this run; empty ≠ none
get_me	context	⭐	403 under integration token — unusable

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	object	1	error	isError envelope only
list_issues	array	3	number, title, state, labels, assignees	Integrity-filtered
list_pull_requests	array	4	number, title, state, body, base, head, user	base+head embed full repo objects
list_discussions	object	3	discussions, pageInfo, totalCount, category	GraphQL-style envelope
get_label	object	1	id, name, color, description	Flat
search_repositories	array	2	full_name, language, stargazers_count, topics	Flat, no URL bloat
get_file_contents	object	3	content, resource, text, uri, sha	Resource wrapper
list_code_scanning_alerts	array	5	number, rule, tool, state, most_recent_instance	Deeply nested, verbose

Response Size Analysis

Toolset	Tokens (today)	Tools Tested
code_security	2,375	list_code_scanning_alerts
repos	615	get_file_contents
pull_requests	460	list_pull_requests
search	125	search_repositories
discussions	110	list_discussions
issues	90	list_issues
context	30	get_me
labels	30	get_label

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	30	object	1	403 Resource not accessible by integration
list_issues	issues	90	array	2	Integrity policy withheld #36242; empty array returned
list_pull_requests	pull_requests	460	array	4	Body verbatim; repo metadata duplicated in base+head
list_discussions	discussions	110	object	5	totalCount=6783, cursor pagination, minimal fields
get_label	labels	30	object	5	color hex + description sufficient
search_repositories	search	125	array	5	minimal_output default; flat discovery signals
get_file_contents	repos	615	object	4	Full file text + SHA for follow-up edits
list_code_scanning_alerts	code_security	2,375	array	2	2 alerts; rule.help markdown dominates payload

30-Day Trend Summary

Metric	Value
Data Points	79
Distinct Days	8 (2026-05-20 → 2026-06-01)
Average Daily Tokens	13,997
Rating Trend	Stable — efficient tools consistently 5/5, code_security consistently heavy

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References:

• §26758897645

Generated by 🔍 GitHub MCP Structural Analysis · opus48 1.4M · ◷

• expires on Jun 2, 2026, 1:55 PM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #36446.