MCP Structural Analysis - 2025-12-24

[github-actions[bot]]

Analyzed 10 GitHub MCP tools across 8 toolsets. Average usefulness rating: 4.1/5. Best performers: get_label (30 tokens, rating 5/5), list_branches (55 tokens, rating 5/5), list_workflows (170 tokens, rating 5/5). Major concern: list_code_scanning_alerts consumes 34,650 tokens with full rule documentation - needs minimal_output option.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	50,380
Average Usefulness Rating	4.1/5
Best Rated Tool	get_file_contents, list_workflows, list_branches, search_repositories, get_label, list_discussions: 5/5
Worst Rated Tool	get_me: 1/5 (403 error), list_code_scanning_alerts: 2/5 (massive bloat)

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
get_file_contents	repos	⭐⭐⭐⭐⭐	Excellent for autonomous agents
list_workflows	actions	⭐⭐⭐⭐⭐	Excellent for autonomous agents
list_branches	repos	⭐⭐⭐⭐⭐	Excellent for autonomous agents
search_repositories	search	⭐⭐⭐⭐⭐	Excellent for autonomous agents
get_label	labels	⭐⭐⭐⭐⭐	Excellent for autonomous agents
list_discussions	discussions	⭐⭐⭐⭐⭐	Excellent for autonomous agents
list_pull_requests	pull_requests	⭐⭐⭐⭐	Good, minor improvements possible
list_issues	issues	⭐⭐⭐⭐	Good, minor improvements possible
list_code_scanning_alerts	code_security	⭐⭐	Limited usefulness
get_me	context	⭐	Poor for agentic tasks

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_file_contents	resource	1	content, sha	Clean file content delivery with SHA. Perfect for agents
list_issues	object_with_array	4	issues, pageInfo, totalCount, number, title, state, body, labels, user	Comprehensive with full body text. Context-heavy due to large bodies
list_pull_requests	array	5	number, title, state, body, user, head, base, assignees, labels	Very detailed with full repo objects. Extremely verbose even with perPage=1
list_workflows	object_with_array	2	total_count, workflows, id, name, path, state, url	Clean and minimal. Perfect balance
list_code_scanning_alerts	array	4	number, rule, tool, state, most_recent_instance	EXTREMELY verbose - includes full help text and examples for every rule
list_discussions	object_with_array	3	discussions, pageInfo, totalCount, number, title, category	Clean with pagination support. Efficient
get_label	object	1	color, description, id, name	Minimal, focused. Most efficient tool tested
search_repositories	object_with_array	3	total_count, incomplete_results, items, name, full_name, description, topics	Efficient with minimal_output default
list_branches	array	1	name, sha, protected	Minimal. Just essential fields
get_me	error	0	-	403 error - not accessible in workflow context

Response Size Analysis

Toolset	Avg Tokens	Tools Tested	Notes
code_security	34,650	1	EXTREMELY verbose - massive bloat from educational content
pull_requests	9,100	1	Very detailed - full repo objects in head/base
issues	4,700	1	Comprehensive - large body text included
get_file_contents	1,100	1	Efficient file delivery
search	420	1	Good balance with minimal_output
actions	170	1	Excellent efficiency
discussions	130	1	Very efficient
repos (branches)	55	1	Minimal and focused
labels	30	1	Most efficient
context	25	1	Error response

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_file_contents	repos	1,100	resource	5	Returns file as MCP resource. Clean README content delivered efficiently. Perfect for agents needing repository file access.
list_issues	issues	4,700	object_with_array	4	Very comprehensive issue data with full body text (5400+ chars). Nested user and label objects. Rich but verbose due to large body content. Excellent structure but context-heavy for discovery.
list_pull_requests	pull_requests	9,100	array	4	Extremely detailed PR data with deep nesting. Full repo objects in head/base with complete metadata. Very long body (3600+ chars). Rich but extremely verbose even with perPage=1. Context-heavy.
list_workflows	actions	170	object_with_array	5	Clean, minimal workflow listing. Essential fields with URL. Perfect efficiency for workflow discovery. Excellent for agents.
list_code_scanning_alerts	code_security	34,650	array	2	EXTREMELY verbose - 138,613 characters total. Includes full rule documentation with examples, recommendations, references for EVERY alert. Massive context overhead from educational content. Core data completely buried in bloat. Desperately needs minimal_output option.
list_discussions	discussions	130	object_with_array	5	Clean structure with pagination. Category info embedded. Title and basic metadata with perPage=1. Excellent efficiency for discovery.
get_label	labels	30	object	5	Minimal label data. Four essential fields. One of most efficient tools tested. Perfect for label operations.
search_repositories	search	420	object_with_array	5	Efficient search with minimal_output default. Essential repo metadata with topics array. Well-structured for discovery. Excellent for agents.
list_branches	repos	55	array	5	Minimal branch data. Three essential fields. Perfect efficiency for branch operations. No bloat.
get_me	context	25	error	1	403 error - not accessible with current permissions. Cannot use for agentic workflows in this context.

30-Day Trend Summary

Metric	Value
Data Points	207 (across 29 days)
Average Daily Tokens	19,569
Average Rating Trend	Stable at 4.0-4.2/5
Days Tracked	Nov 26 - Dec 24 (29 days)

Recommendations

Based on the analysis:

High-value tools (rating 4-5):

• get_label - Most efficient (30 tokens) with perfect rating
• list_branches - Minimal (55 tokens), essential fields only
• list_workflows - Excellent balance (170 tokens)
• list_discussions - Very efficient (130 tokens)
• search_repositories - Good efficiency (420 tokens) with minimal_output
• get_file_contents - Clean file delivery (1,100 tokens)
• list_issues - Comprehensive but large bodies (4,700 tokens)
• list_pull_requests - Rich but verbose (9,100 tokens)

Tools needing improvement:

• list_code_scanning_alerts - Desperately needs minimal_output option to strip educational content. Current 34,650 tokens is unusable for agents doing discovery. Core alert data could be <500 tokens without help text.
• get_me - Inaccessible with GitHub App token (403 error). Not useful in workflow context.

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

• get_label (30 tokens, 5/5)
• list_branches (55 tokens, 5/5)
• list_discussions (130 tokens, 5/5)
• list_workflows (170 tokens, 5/5)
• search_repositories (420 tokens, 5/5)

Context-heavy tools (high tokens):

• list_code_scanning_alerts (34,650 tokens, 2/5) - ⚠️ CRITICAL ISSUE
• list_pull_requests (9,100 tokens, 4/5) - Consider body truncation option
• list_issues (4,700 tokens, 4/5) - Body can be very long

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Key Insights

1. Most MCP tools are excellent for agents - 6 out of 10 tools rated 5/5, 2 rated 4/5
2. Efficiency sweet spot - Tools under 500 tokens with rating 5/5 are ideal for agentic discovery
3. Critical issue with code_security toolset - list_code_scanning_alerts bloated with educational content (138KB response). Needs urgent minimal_output option.
4. Body text drives token usage - Issues and PRs would benefit from optional body truncation (first 500 chars) or body_preview field
5. Shallow nesting is better - Tools with depth 1-2 (labels, branches, workflows) are most efficient and easiest for agents to parse
6. MCP resource format works well - get_file_contents delivering files as MCP resources is clean and immediately actionable

References:

• §20484771263

AI generated by GitHub MCP Structural Analysis