[mcp-analysis] MCP Structural Analysis - 2026-04-15

[github-actions[bot]]

Analysis of 9 GitHub MCP tools across 9 toolsets on 2026-04-15. Total context cost: 47,935 tokens. Average usefulness rating: 3.11/5. Two toolsets (discussions, pull_requests) scored excellent (5/5); two (code_security, search) scored poor (≤2/5) due to massive response bloat.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	9
Total Tokens (Today)	47,935
Average Usefulness Rating	3.11/5
Best Rated Tools	list_discussions & list_pull_requests: 5/5
Worst Rated Tool	get_me (403 error) & search_code: 1–2/5
Most Context-Heavy	list_code_scanning_alerts: ~35,800 tokens
Most Context-Efficient	list_discussions: ~105 tokens

---

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
list_discussions	discussions	⭐⭐⭐⭐⭐	Compact, paginated, category-filterable — ideal for agents
list_pull_requests	pull_requests	⭐⭐⭐⭐⭐	Rich structure: head/base/assignees/draft — immediately actionable
list_issues	issues	⭐⭐⭐⭐	Full issue data with body; body length unpredictable
get_file_contents	repos	⭐⭐⭐	Useful but no size guard — large files spike context heavily
list_workflows	actions	⭐⭐⭐	Ignores per_page=1; returns 30 workflows; URL fields are noisy
list_label	labels	⭐⭐⭐	Returns all 100+ labels at once; no effective pagination
list_code_scanning_alerts	code_security	⭐⭐	143k-char response exceeded max tokens; deeply nested; no page limit
search_code	search	⭐⭐	Per-result repo sub-object bloated with 30+ URL templates
get_me	context	⭐	403 forbidden in this environment; not usable

---

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	error	1	error	Not accessible
get_file_contents	object	2	content, sha, mimeType, uri	Full file text inline — no streaming
list_issues	array	3	number, title, state, body, user	Body can be arbitrarily long
list_pull_requests	array	4	number, title, state, head, base, assignees, draft	head/base are nested repo objects
list_workflows	object	2	id, name, path, state, url, html_url	total_count useful; many redundant URL fields
list_code_scanning_alerts	array	5	number, rule, severity, location, instance	Deeply nested; no per_page enforcement
list_discussions	object	3	number, title, category, user, created_at	Cursor pagination; clean schema
list_label	object	2	name, color, description, id	GraphQL node IDs add bloat; all labels returned
search_code	object	4	name, path, sha, html_url, repository	repo sub-object has 30+ URL template fields

---

Response Size Analysis

Toolset	Tokens	Notes
code_security	~35,800	Exceeds max token limit entirely
repos	~4,200	Scales with file size
actions	~3,600	Fixed 30-item response regardless of per_page
labels	~2,250	All labels returned at once
search	~830	Low count but bloated repo sub-object per result
issues	~820	Body-dependent; can spike
pull_requests	~310	Efficient for data density
discussions	~105	Most context-efficient tool tested
context	~20	Error only

---

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	20	error	1	403 in GH Actions environment
get_file_contents	repos	4,200	object/2	3	Large files = large context; README.md ~16KB
list_issues	issues	820	array/3	4	1 issue with full body; body is 500+ words
list_pull_requests	pull_requests	310	array/4	5	Rich, compact, actionable
list_workflows	actions	3,600	object/2	3	30 workflows returned; per_page ignored
list_code_scanning_alerts	code_security	35,800	array/5	2	143k chars; overflows tool result buffer
list_discussions	discussions	105	object/3	5	Best ratio of info to tokens
list_label	labels	2,250	object/2	3	100+ labels; no page control observed
search_code	search	830	object/4	2	Repo sub-object is ~80% of token cost

---

30-Day Trend Summary

Metric	Value
Data Points	9 (first run)
Total Tokens Today	47,935
Average Usefulness	3.11/5
Trend	Baseline established — will track changes over 30 days

---

Recommendations

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

• list_discussions — most efficient for discovery
• list_pull_requests — rich PR metadata in compact form
• list_issues — good but set low perPage

Tools needing improvement:

• list_code_scanning_alerts — must enforce pagination; response size is unacceptable
• search_code — strip redundant URL templates from embedded repo sub-object
• list_workflows — honor per_page parameter
• list_label — add proper pagination support

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

• list_discussions (105 tokens, rating 5)
• list_pull_requests (310 tokens, rating 5)

Context-heavy tools to use sparingly:

• list_code_scanning_alerts — use with extreme perPage limits
• list_workflows — filter by specific workflow name when possible
• get_file_contents — only on small, known files

Visualizations

Response Size by Toolset

Usefulness Ratings

Token Usage by Tool (2026-04-15)

Size vs Usefulness

References:

• §24453664185

Generated by GitHub MCP Structural Analysis · ● 326.5K · ◷

• expires on Apr 16, 2026, 12:17 PM UTC