[mcp-analysis] MCP Structural Analysis - November 26, 2025

[github-actions[bot]]

This analysis evaluates GitHub MCP tools for both response size (token efficiency) and structural usefulness for autonomous agents. 10 representative tools were tested across 8 toolsets. Key findings: Average usefulness rating of 4.78/5, with most tools providing excellent structure for agentic work. Notable: get_label (65 tokens, 5/5) and list_branches (95 tokens, 5/5) are exceptionally efficient. list_pull_requests is context-heavy at 3,200 tokens but still highly useful (4/5).

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Valid Tools (non-error)	9
Total Tokens (Today)	7,900
Average Tokens	878
Average Usefulness Rating	4.78/5
Best Rated Tool	7 tools at 5/5 (see below)
Most Efficient Tool	get_label: 65 tokens, 5/5
Most Context-Heavy	list_pull_requests: 3,200 tokens, 4/5

Usefulness Ratings for Agentic Work

Ratings are based on five criteria: Completeness, Actionability, Clarity, Efficiency, and Relationships.

Tool	Toolset	Tokens	Rating	Assessment
get_label	labels	65	⭐⭐⭐⭐⭐	Minimal, focused - perfect for label ops
list_branches	repos	95	⭐⭐⭐⭐⭐	Essential branch data, no bloat
list_discussions	discussions	240	⭐⭐⭐⭐⭐	Clean with pagination, category embedded
list_workflows	actions	280	⭐⭐⭐⭐⭐	Perfect balance of completeness/efficiency
list_commits	repos	420	⭐⭐⭐⭐⭐	Well-structured, logical nesting
list_issues	issues	850	⭐⭐⭐⭐⭐	Comprehensive with all context included
get_file_contents	repos	1,350	⭐⭐⭐⭐⭐	Clean resource format, immediately actionable
search_code	search	1,400	⭐⭐⭐⭐	Rich but context-heavy, deep nesting
list_pull_requests	pull_requests	3,200	⭐⭐⭐⭐	Very detailed but verbose, some redundancy
get_me	context	25	⭐	403 error - not accessible in this environment

Schema Analysis

Tool	Schema Type	Depth	Key Fields	Notes
get_label	object	1	color, description, id, name	Flat structure, optimal
list_branches	array	1	name, sha, protected	Simple array, minimal depth
list_workflows	object+array	2	total_count, workflows, id, name, path	Logical grouping
list_discussions	object+array	3	discussions, pageInfo, number, title, category	Pagination included
list_commits	array	3	sha, commit, author, committer, message	Nested author/committer objects
list_issues	object+array	4	issues, pageInfo, number, title, state, labels, user	Rich with embedded entities
search_code	object+array	5	total_count, items, name, path, repository	Deep due to full repo context
list_pull_requests	array	5	number, title, state, user, head, base, assignees	Very deep nesting, full PR context
get_file_contents	resource	1	content, sha, encoding	MCP resource format
get_me	error	0	N/A	Permission denied

Response Size Analysis

Toolset	Avg Tokens	Tools Tested	Efficiency Rating
labels	65	1	Excellent - minimal size
repos	655	3	Good - balanced
discussions	240	1	Excellent - compact
actions	280	1	Excellent - efficient
issues	850	1	Good - comprehensive
search	1,400	1	Moderate - context-rich
pull_requests	3,200	1	Heavy - very detailed
context	N/A	0 (error)	Not accessible

Tool-by-Tool Analysis

Excellent (5/5) - Highly Recommended for Agents

get_label (labels, 65 tokens)

• Why 5/5: Minimal, focused data. Returns exactly what's needed for label operations with no bloat.
• Best for: Label management, issue/PR tagging operations

list_branches (repos, 95 tokens)

• Why 5/5: Essential branch data (name, SHA, protected status) in a simple array. Perfect efficiency.
• Best for: Branch listing, protection checks, branch-based workflows

list_discussions (discussions, 240 tokens)

• Why 5/5: Clean structure with pagination support and category info embedded. All needed context.
• Best for: Discussion analysis, community engagement tracking

list_workflows (actions, 280 tokens)

• Why 5/5: Perfect balance of completeness and efficiency. Total count + workflow details.
• Best for: Workflow discovery, CI/CD analysis

list_commits (repos, 420 tokens)

• Why 5/5: Well-structured with logical nesting. Author/committer objects make sense.
• Best for: Commit history analysis, authorship tracking

list_issues (issues, 850 tokens)

• Why 5/5: Comprehensive issue data with pagination. All context included (labels, user, state).
• Best for: Issue tracking, triage, trend analysis

get_file_contents (repos, 1,350 tokens)

• Why 5/5: Returns file content as MCP resource. Clean, immediately actionable format.
• Best for: File reading, content analysis, code inspection

Good (4/5) - Useful with Minor Trade-offs

search_code (search, 1,400 tokens)

• Why 4/5: Rich search results with full repository context. Deep nesting (level 5) due to embedded repo objects. Useful but context-heavy.
• Trade-off: Includes full repo metadata which may be redundant for simple searches.
• Best for: Cross-repo code search, pattern discovery

list_pull_requests (pull_requests, 3,200 tokens)

• Why 4/5: Very detailed PR data with deep nesting (level 5). Includes repo objects, user objects, head/base details. Rich but verbose.
• Trade-off: Some fields redundant for basic agentic tasks. Consider using perPage=1 to reduce context.
• Best for: Comprehensive PR analysis, detailed code review workflows

Limited (1/5) - Not Usable

get_me (context, error)

• Why 1/5: Returns 403 error - not accessible with current GitHub Actions permissions.
• Workaround: Use environment variables or repository context instead.

30-Day Trend Summary

Metric	Value
Data Points	10 (first day of tracking)
Date Range	2025-11-26 (single day)
Historical Data	Starting fresh - future runs will show trends

Recommendations

High-Value Tools (Rating 4-5)

These tools provide excellent structure for autonomous agents:

• Ultra-efficient: get_label, list_branches, list_discussions
• Well-balanced: list_workflows, list_commits, list_issues
• Content-rich: get_file_contents, search_code, list_pull_requests

Context-Efficient Tools (Low tokens, high rating)

Best for context-constrained agents:

1. get_label - 65 tokens, 5/5
2. list_branches - 95 tokens, 5/5
3. list_discussions - 240 tokens, 5/5

Context-Heavy Tools (High tokens)

Use with pagination or minimal parameters:

1. list_pull_requests - 3,200 tokens (consider perPage=1)
2. search_code - 1,400 tokens
3. get_file_contents - 1,350 tokens (varies by file size)

Schema Design Patterns

Best practices observed:

• Flat structures (depth 1-2): Most efficient for agents
• Pagination support: Essential for scalability (pageInfo, total_count)
• Embedded relationships: Reduces need for follow-up calls
• Minimal redundancy: Tools like get_label show optimal design

Areas for improvement:

• list_pull_requests: Consider a "minimal" mode with reduced repo metadata
• search_code: Option to exclude full repository objects

Visualizations

Response Size by Toolset

Shows average response size across different toolsets. Pull requests toolset is notably larger.

Usefulness Ratings

All toolsets rate highly (4-5 stars) for agentic usefulness. Green bars (≥4) indicate excellent tools.

Size vs Usefulness

Scatter plot showing most tools cluster in the "high value, reasonable size" quadrant. list_pull_requests is an outlier (large but useful).

Response Size by Tool

Horizontal bar chart showing size distribution. Colors indicate usefulness: green (excellent), orange (adequate), red (limited).

Conclusion

The GitHub MCP toolset demonstrates excellent design for agentic workflows, with an average usefulness rating of 4.78/5. Most tools strike a strong balance between completeness and efficiency. Key strengths:

• Clear schemas with logical nesting
• Pagination support for scalability
• Embedded relationships reduce API calls
• Minimal redundancy in most tools

For future iterations, consider:

• Tracking response size trends over 30 days
• Adding "minimal" modes for verbose endpoints
• Testing additional toolsets (security, releases)

AI generated by GitHub MCP Structural Analysis

[pelikhan]

/q

• explore a bit more the data, use the shared agentic workflow that pre download data
• show compact preview of data structure
• look for duplicate field that waste token and could be inferred

[github-actions[bot]]

Agentic Q triggered by this discussion comment.

[github-actions[bot]]

⚓ Avast! This discussion be marked as outdated by GitHub MCP Structural Analysis.
🗺️ A newer treasure map awaits ye at Discussion #5289.
Fair winds, matey! 🏴‍☠️