[mcp-analysis] MCP Structural Analysis - 2026-02-10

[github-actions[bot]]

Analyzed 12 GitHub MCP tools for response structure and usefulness. Average rating: 4.00/5 — most tools are well-designed for agentic work. Key findings: security tools return massive payloads (95KB), some REST endpoints ignore pagination, while GraphQL-based tools deliver optimal structure. Three tools flagged for improvement.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	12
Total Tokens (Today)	37,725
Average Usefulness Rating	4.00/5
Best Rated Tool	get_file_contents: 5/5
Worst Rated Tool	get_me: 1/5
Context-Efficient Tools	7 tools under 500 tokens
Oversized Responses	3 tools exceeded MCP limits

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
get_file_contents	repos	⭐⭐⭐⭐⭐	Excellent - minimal overhead, just content
list_issues	issues	⭐⭐⭐⭐⭐	Excellent - rich GraphQL structure with full metadata
list_discussions	discussions	⭐⭐⭐⭐⭐	Excellent - clean GraphQL with cursor pagination
list_branches	repos	⭐⭐⭐⭐⭐	Excellent - minimal array, essential data only
search_repositories	search	⭐⭐⭐⭐⭐	Excellent - comprehensive metadata for discovery
list_commits	repos	⭐⭐⭐⭐⭐	Excellent - balanced detail for commit analysis
get_label	labels	⭐⭐⭐⭐⭐	Excellent - minimal, complete label metadata
search_issues	search	⭐⭐⭐⭐	Good - full issue bodies, some user object redundancy
list_pull_requests	pull_requests	⭐⭐⭐	Adequate - 18KB payload with duplicated repo objects
list_workflows	actions	⭐⭐⭐	Adequate - pagination broken, returns all 192 workflows
list_code_scanning_alerts	security	⭐⭐	Limited - 95KB payload, no pagination, hard to parse
get_me	context	⭐	Poor - 403 error in GitHub Actions, unavailable

Schema Analysis

Tool	Type	Depth	Key Fields	Structure Notes
get_file_contents	string	1	content	Simple text response
list_branches	array	2	name, sha, protected	Minimal array
get_label	object	1	id, name, color, description	Flat object
list_discussions	object	3	discussions[], pageInfo, totalCount	GraphQL with cursor pagination
search_repositories	object	3	total_count, items[].name, items[].stargazers_count	Search result wrapper
list_commits	array	3	sha, commit.message, author.login	Nested author info
list_issues	object	4	issues[], pageInfo, totalCount, full body	GraphQL with rich metadata
search_issues	object	4	total_count, items[].labels, items[].user	Full issue + user objects
list_pull_requests	object	5	number, title, head.repo, base.repo	Deep nesting, full repo duplication
list_workflows	object	2	workflows[].id, workflows[].name	Flat but oversized array
list_code_scanning_alerts	object	5	number, rule, tool, locations	Extremely nested with full rule descriptions
get_me	error	0	-	Permission error

Response Size Analysis

Toolset	Avg Tokens	Tools Tested	Efficiency
security	23,800	1	❌ Very Heavy
pull_requests	4,650	1	⚠️ Heavy
actions	3,900	1	⚠️ Heavy
search	1,200	3	✅ Moderate
issues	1,650	1	✅ Moderate
repos	125	4	✅ Light
discussions	275	1	✅ Light
labels	50	1	✅ Very Light
context	0	1	❌ Unavailable

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_file_contents	repos	25	string	⭐⭐⭐⭐⭐	Clean text response with minimal overhead. Just returns file contents. Perfect for agents - efficient and actionable.
list_branches	repos	100	array	⭐⭐⭐⭐⭐	Minimal, clean array response. Just essential data: name, SHA, protection status. Highly efficient for agents.
get_label	labels	50	object	⭐⭐⭐⭐⭐	Minimal response with just label metadata. Efficient and complete for label operations.
list_discussions	discussions	275	object	⭐⭐⭐⭐⭐	GraphQL structure with cursor pagination. Clean response with category, user, timestamps. Well-organized and efficient for agents.
list_commits	repos	325	array	⭐⭐⭐⭐⭐	Clean commit history with message, author, committer. Good balance of detail. Efficient for commit analysis tasks.
search_repositories	search	350	object	⭐⭐⭐⭐⭐	Excellent search results with rich metadata. Includes stars, topics, language, dates. All needed discovery data present. Immediately actionable.
list_issues	issues	1,650	object	⭐⭐⭐⭐⭐	GraphQL response with cursor pagination. Rich issue details including full body, labels, user info. Excellent structure for agents. Complete and actionable.
search_issues	search	2,600	object	⭐⭐⭐⭐	Comprehensive search results with full issue bodies. Includes extensive user info with many URL fields. Some redundancy in user object but overall good. Actionable for finding specific issues.
list_workflows	actions	3,900	oversized	⭐⭐⭐	Response too large (15KB) for single workflow with perPage=1. Returns full list of 192 workflows regardless of perPage parameter. Pagination appears broken. Usable but inefficient.
list_pull_requests	pull_requests	4,650	oversized	⭐⭐⭐	Response too large (18KB) for MCP, requires file system read. Includes full repo objects for both head and base with ALL fields duplicated. Heavy redundancy. Usable but context-inefficient.
list_code_scanning_alerts	security	23,800	oversized	⭐⭐	Massive payload (95KB) for security alerts. Returns extensive alert details with full rule descriptions, locations, and analysis. No pagination respected. Context-heavy and hard to process. Needs filtering options.
get_me	context	0	error	⭐	403 error - not accessible by integration. Tool is unavailable in GitHub Actions context, making it useless for agentic workflows running in CI.

30-Day Trend Summary

Metric	Value
Data Points	26 records (2 analysis runs)
Date Range	Feb 6 - Feb 10, 2026
Tools Tracked	12 unique tools
Average Daily Tokens	~38,000 tokens per analysis
Rating Trend	Stable at 4.00-4.07 average

Notable Changes:

• list_issues token count increased from 650 → 1,650 (154% increase) due to richer issue body content in latest analysis
• list_pull_requests token count increased from 2,800 → 4,650 (66% increase) with full repo duplication
• list_workflows token count increased from 3,200 → 3,900 (22% increase)
• list_code_scanning_alerts saw massive increase from unmeasured → 23,800 tokens on first full test

Recommendations

🟢 High-Value Tools (Rating 4-5, recommend for frequent use):

• get_file_contents, list_branches, get_label — ultra-efficient, minimal context
• list_issues, list_discussions — excellent GraphQL structure
• search_repositories, list_commits — well-balanced metadata

🟡 Use with Caution (Rating 3, context-heavy):

• list_pull_requests — consider using minimal fields or get single PR instead of list
• list_workflows — pagination broken, consider filtering by name
• search_issues — prefer list_issues with filters when possible

🔴 Needs Improvement (Rating 1-2):

• list_code_scanning_alerts — add filtering parameters, respect pagination, reduce payload size
• get_me — fix permissions for GitHub Actions integration context

💡 Context-Efficient Alternatives:

• Instead of list_pull_requests → use search_pull_requests with minimal fields
• Instead of list_code_scanning_alerts → add severity/state filters to reduce payload
• Instead of list_workflows → use get_workflow when workflow name is known

Visualizations

Response Size by Toolset

Analysis: Security toolset dominates with 23,800 tokens average. Actions and pull_requests toolsets are also heavy (3,900-4,650 tokens). Repos, discussions, and labels toolsets are highly efficient (<500 tokens average).

Usefulness Ratings by Toolset

Analysis: Most toolsets score 4+ (good to excellent). Security and context toolsets score lower due to oversized payloads and permission errors respectively. Green bars (repos, discussions, labels, issues, search) indicate excellent agentic workflow support.

Daily Token Trend (30-Day Window)

Analysis: Token usage increased significantly from Feb 6 (12,750 tokens) to Feb 10 (37,725 tokens) — a 196% increase. This reflects deeper testing including security alerts and larger issue/PR payloads. Trend suggests response sizes are growing as more complete data is returned.

Token Size vs Usefulness Scatter

Analysis: Sweet spot is top-left quadrant (low tokens, high usefulness). Most repos, search, discussions, and labels tools cluster here. Security and pull_requests tools sit in the bottom-right (high tokens, lower usefulness). Ideal tools for agents are those rated 5 with <500 tokens.

References: §21862776279

AI generated by GitHub MCP Structural Analysis

• expires on Feb 17, 2026, 11:25 AM UTC