[mcp-analysis] MCP Structural Analysis - 2026-03-23

[github-actions[bot]]

10 tools analyzed across 9 toolsets today. Average usefulness: 3.30/5. Top performers: get_label, search_repositories, list_discussions (all 5/5). Persistent outlier: list_code_scanning_alerts (1/5, ~137KB, no pagination). The list_workflows tool continues to ignore per_page parameter. 30-day dataset now at 210 records across 21 measurement days.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	37,735
Avg Usefulness Rating	3.30/5
Best Rated Tool	get_label, search_repositories, list_discussions: 5/5
Worst Rated Tool	get_me: 1/5 (DIFC policy block), list_code_scanning_alerts: 1/5 (size)
30-day Avg Usefulness	2.90/5
30-day Avg Daily Tokens	30,476

---

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
get_label	labels	⭐⭐⭐⭐⭐	Flat object, ~25 tokens, all needed fields, consistently #1
search_repositories	search	⭐⭐⭐⭐⭐	Rich metadata, topics, stats, minimal_output prevents bloat
list_discussions	discussions	⭐⭐⭐⭐⭐	Compact, cursor pagination, perPage respected, 75 tokens
list_commits	repos	⭐⭐⭐⭐	Clean nested structure, perPage respected, actionable
list_pull_requests	pull_requests	⭐⭐⭐⭐	Good schema, perPage respected, draft flag present
get_file_contents	repos	⭐⭐⭐	Full content returned but verbose wrapper; large files bloat context
list_issues	issues	⭐⭐⭐	Blocked by DIFC policy today; typically good when accessible
list_workflows	actions	⭐⭐	Ignores per_page (returns 30), 3 redundant URL fields each
get_me	context	⭐	Empty response — DIFC integrity policy blocks user-scoped data
list_code_scanning_alerts	code_security	⭐	~137KB response, no per_page support, 5 levels deep, unusable inline

---

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	array	0	—	Empty — DIFC policy
get_file_contents	object	3	sha, content[].resource.text, .uri	Resource wrapper adds overhead
list_issues	array	0	—	DIFC-blocked today
list_pull_requests	array	4	number, title, state, body, head.ref, base.ref	Body may contain HTML-encoded text
list_workflows	object	2	id, name, path, state, url, html_url, badge_url	3 URL fields per workflow is redundant
list_discussions	object	3	number, title, category.name, user.login, pageInfo	Excellent cursor pagination support
get_label	object	1	id, name, color, description	Optimal flat structure
search_repositories	array	2	full_name, language, topics[], stargazers_count	Topics as array is very useful
list_commits	array	3	sha, commit.message, commit.author, author.login	Well-structured, no bloat
list_code_scanning_alerts	array	5	number, rule.id, rule.severity, location.path	Deeply nested, no size control

---

Response Size Analysis

Toolset	Avg Tokens	Notes
code_security	~34,000	Consistent outlier — no per_page support
repos	~700	Driven by get_file_contents file size
actions	~1,750	per_page ignored, returns 30 workflows
pull_requests	~400–2,500	Varies with PR body length
issues	~5–1,250	When accessible, body can be large
discussions	~75	Most efficient paginated list
search	~100	Excellent efficiency
labels	~25	Best-in-class minimal response
commits	~175	Clean and efficient
context	~5	Non-functional in this integration

---

Tool-by-Tool Analysis (Today)

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	5	array/empty	1/5	DIFC policy block
get_file_contents	repos	1,250	object/3	3/5	README ~6K chars; bloated for large files
list_issues	issues	5	array/0	3/5	1 item removed by DIFC policy
list_pull_requests	pull_requests	375	array/4	4/5	perPage=1 respected; good structure
list_workflows	actions	1,750	object/2	2/5	per_page ignored; 30 returned of 234
list_discussions	discussions	75	object/3	5/5	Compact, paginated, totalCount=4037
get_label	labels	25	object/1	5/5	Minimal, complete, flat
search_repositories	search	100	array/2	5/5	minimal_output=true works well
list_commits	repos	150	array/3	4/5	Clean, perPage=1 respected
list_code_scanning_alerts	code_security	34,000	array/5	1/5	Historical value; ~137KB, no pagination

---

30-Day Trend Summary

Metric	Value
Data Points	210
Measurement Days	21
Avg Daily Tokens (all tools)	30,476
Avg Daily Tokens (excl. code_security)	~3,600
Rating Trend	Stable (2.85–3.30 range)
Most Consistent Performer	get_label (5/5 every day)
Most Consistent Underperformer	list_code_scanning_alerts (1/5 every day)

---

Recommendations

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

• get_label, search_repositories, list_discussions — always efficient and complete
• list_commits, list_pull_requests — good structure, perPage respected

Tools needing improvement:

• list_workflows — must respect per_page parameter; remove badge_url (rarely useful for agents)
• list_code_scanning_alerts — needs per_page support; deeply nested structure hinders parsing
• get_file_contents — content wrapper is verbose; consider streaming for large files

Context-efficient sweet spots (low tokens, high rating):

• get_label (25 tokens, 5/5), search_repositories (100 tokens, 5/5), list_discussions (75 tokens, 5/5)

Avoid in context-constrained agents:

• list_code_scanning_alerts — always triggers file-based payload; use only when explicitly needed
• get_file_contents on large files — README alone is 1,250 tokens

DIFC / Integration Note:

• get_me is permanently blocked by integrity policy in this integration — use get_repository or context from workflow env instead

---

Visualizations

Response Size by Toolset

Usefulness Ratings (Today)

Daily Token Usage Trend (30 days)

Token Size vs Usefulness (30-day avg)

Usefulness Rating Trends by Tool

References: §23434630376

AI generated by GitHub MCP Structural Analysis · history

• expires on Mar 24, 2026, 11:22 AM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #22654.