[mcp-analysis] MCP Structural Analysis - 2026-05-22

[github-actions[bot]]

Overview

Analyzed 10 representative GitHub MCP tools across 9 toolsets. Today's run measured 25,560 total tokens with an average usefulness rating of 3.5/5. The standout performers remain search_repositories (concise, 5/5) and get_file_contents (complete, 5/5); the persistent laggards are get_me (blocked by integration permissions, 1/5) and list_label (no pagination control, 2/5).

Today's Snapshot

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	25,560
Average Usefulness Rating	3.5/5
Best Rated Tools	get_file_contents, search_repositories (5/5)
Worst Rated Tool	get_me (1/5)
Data Points (30-day window)	30

Key Recommendations

• Highest-value tools (rating 4–5, use freely): search_repositories, get_file_contents, list_pull_requests, list_issues, list_code_scanning_alerts, list_discussions.
• Context-efficient sweet spot (low tokens + high rating): search_repositories (~130 tokens, 5/5), list_discussions (~200 tokens, 4/5).
• Context-heavy tools — paginate carefully: list_workflows (~5.5K), search_issues (~4.8K), list_issues (~4.5K), get_file_contents (~4.2K), list_label (~3.6K).
• Avoid in this workflow context: get_me consistently returns 403 — skip in workflows using a constrained integration token.
• Wish-list improvement: list_label should respect a perPage parameter; currently always returns 100 of 570 labels.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Toolsets Covered	context, repos, issues, pull_requests, actions, code_security, discussions, labels, search
Total Tokens (Today)	25,560
Average Usefulness Rating	3.5/5
Best Rated Tool	get_file_contents: 5/5
Worst Rated Tool	get_me: 1/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
get_file_contents	repos	⭐⭐⭐⭐⭐	Excellent — full file content + SHA + URI in one call
search_repositories	search	⭐⭐⭐⭐⭐	Excellent — minimal_output keeps responses small and actionable
list_issues	issues	⭐⭐⭐⭐	Good — body, labels, user.login all present; body can dominate size
list_pull_requests	pull_requests	⭐⭐⭐⭐	Good — base/head/assignees included; some duplication in base+head
list_code_scanning_alerts	code_security	⭐⭐⭐⭐	Good — actionable rule + location; rule.help is verbose
list_discussions	discussions	⭐⭐⭐⭐	Good — cursor pagination + totalCount; needs follow-up for body
list_workflows	actions	⭐⭐⭐	Adequate — perPage=1 not honored strictly; includes badge URLs
search_issues	search	⭐⭐⭐	Adequate — full body + many *_url fields inflate size
list_label	labels	⭐⭐	Limited — no per-call pagination; always returns 100 labels
get_me	context	⭐	Poor — 403 in this integration; never returns useful data here

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	object	2	content, isError	Error wrapper only in this env
get_file_contents	resource	3	content.text, SHA, uri, mimeType	Returns text + identity
list_issues	array	3	number, title, state, body, labels, user, timestamps	Body is full markdown
list_pull_requests	array	4	number, title, base, head, body, user, assignees, draft, merged	Nested repo objects in base/head
list_workflows	object	3	total_count, workflows[].id/name/path/state/html_url	Wrapped object with total_count
list_code_scanning_alerts	array	4	number, rule, tool, state, most_recent_instance.location	Nested instance + rule.help markdown
list_discussions	object	3	discussions[], pageInfo, totalCount	GraphQL-style envelope
list_label	object	2	labels[].id/name/color/description, totalCount	Flat list, no per-call pagination
search_repositories	array	2	full_name, description, language, stars, forks, topics, html_url	Minimal, no fluff
search_issues	array	4	number, title, body, labels, user, reactions, author_association, many *_url fields	Many URL fields bloat output

Response Size Analysis (today)

Toolset	Tokens (approx)
actions	5,500
issues	4,500
search (search_issues)	4,800
repos	4,200
labels	3,600
code_security	1,400
pull_requests	1,200
discussions	200
search (search_repositories)	130
context	30 (error)

Tool-by-Tool Notes

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	30	object	1	403 in workflow integration; consistent across 3 days
get_file_contents	repos	4,200	resource	5	Full README + SHA returned; no size cap could bloat for large files
list_issues	issues	4,500	array	4	Full body included; auto-generated reports dominate the response
list_pull_requests	pull_requests	1,200	array	4	Rich metadata; nested repo duplicated in base+head
list_workflows	actions	5,500	object	3	perPage=1 returned ~30 workflows; many fields rarely useful
list_code_scanning_alerts	code_security	1,400	array	4	Verbose rule.help markdown inflates per-alert size
list_discussions	discussions	200	object	4	Compact GraphQL envelope; body not included
list_label	labels	3,600	object	2	Always returns 100 of 570 labels; bloated for simple lookups
search_repositories	search	130	array	5	minimal_output is best-in-class for context efficiency
search_issues	search	4,800	array	3	Full body + many *_url fields; high cost per item

30-Day Trend Summary

Metric	Value
Data Points	30
Days Covered	3 (2026-05-20, 2026-05-21, 2026-05-22)
Average Tokens per Tool (today)	2,556
Average Rating Trend	Stable around 3.4–3.5

Rating distribution has stayed consistent across the three days. get_me is the only persistent 1/5, and search_repositories / get_file_contents are the persistent 5/5 anchors. list_label newly surfaced a totalCount field on 2026-05-21 (vs. 2026-05-20) but its 2/5 rating is unchanged because the pagination issue remains.

Visualizations

Response Size by Toolset (30-day average)

Usefulness Ratings by Toolset (today)

Daily Total Token Trend

Size vs Usefulness Scatter (today)

References:

• §26288220700

Generated by 🔍 GitHub MCP Structural Analysis · ● 13.7M · ◷

• expires on May 23, 2026, 12:46 PM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-05-23T12:46:44.129Z.

Closed by Workflow