[mcp-analysis] MCP Structural Analysis - 2026-02-05

[github-actions[bot]]

Analyzed 10 GitHub MCP tools for response size (tokens), schema structure, and usefulness for autonomous agents. Key findings: Best tools average 4.2/5 rating with clean schemas (≤2 nesting depth). Worst: list_pull_requests (3/5, 4800 tokens, depth 5). Most efficient: list_branches (5/5, 200 tokens). Critical issue: get_me returns 403 error, blocking agent access to user context.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	15,050
Average Token Size	1,505
Average Usefulness Rating	3.8/5
Best Rated Tools	get_file_contents, search_repositories, list_branches: 5/5
Worst Rated Tool	get_me: 1/5 (403 error)
Most Context-Heavy	list_pull_requests: 4,800 tokens
Most Efficient	list_branches: 200 tokens

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Tokens	Assessment
get_file_contents	repos	⭐⭐⭐⭐⭐	50	Excellent - Clean string response, minimal context, immediately actionable
search_repositories	search	⭐⭐⭐⭐⭐	1,800	Excellent - Rich metadata without bloat, perfect for discovery
list_branches	branches	⭐⭐⭐⭐⭐	200	Excellent - Minimal, efficient, exactly what agents need
list_issues	issues	⭐⭐⭐⭐	1,200	Good - Complete data with embedded objects, slight nesting complexity
list_workflows	actions	⭐⭐⭐⭐	2,400	Good - Well-structured, reasonable context usage
list_discussions	discussions	⭐⭐⭐⭐	600	Good - Concise with category info, efficient scanning
list_commits	commits	⭐⭐⭐⭐	800	Good - Clean commit data, balanced detail
list_pull_requests	pull_requests	⭐⭐⭐	4,800	Adequate - Very verbose, deep nesting (5 levels), redundant URLs
list_label	labels	⭐⭐⭐	3,200	Adequate - Returns ALL labels (353) without pagination control
get_me	context	⭐	0	Poor - 403 error, blocks agent access to user context

Schema Analysis

Tool	Type	Depth	Key Fields	Complexity
get_file_contents	string	1	SHA, content	Simple ✅
list_branches	array	2	name, sha, protected	Simple ✅
list_discussions	object	3	discussions, pageInfo, number, title, category	Moderate
list_commits	array	3	sha, commit, message, author	Moderate
list_issues	object	4	issues, pageInfo, number, title, labels, user	Moderate
search_repositories	object	4	total_count, items, name, description, topics	Moderate
list_workflows	object	2	total_count, workflows, id, name, path	Simple ✅
list_label	object	2	labels, totalCount, name, color	Simple ✅
list_pull_requests	array	5	number, title, head, base, _links	Complex ⚠️
get_me	error	0	N/A	Error ❌

Response Size Analysis

Toolset	Avg Tokens	Tools Tested	Efficiency
pull_requests	4,800	1	Heavy ⚠️
labels	3,200	1	Heavy ⚠️
actions	2,400	1	Moderate
search	1,800	1	Moderate
issues	1,200	1	Moderate
commits	800	1	Light ✅
discussions	600	1	Light ✅
branches	200	1	Light ✅
repos	50	1	Light ✅
context	0	1	Error ❌

Tool-by-Tool Analysis

⭐⭐⭐⭐⭐ Excellent Tools (Rating 5/5)

get_file_contents (repos)

• Tokens: 50
• Schema: Simple string with SHA
• Nesting: 1 level
• Why Excellent: Clean string response with file content. Perfect for agents - simple, actionable, efficient. Minimal context usage.
• Recommendation: Ideal for file reading tasks.

search_repositories (search)

• Tokens: 1,800
• Schema: Object with items array
• Nesting: 4 levels
• Why Excellent: Rich metadata (name, description, topics, stars) without bloat. Uses minimal_output mode effectively. Perfect for discovery tasks.
• Recommendation: Use for finding repos by criteria.

list_branches (branches)

• Tokens: 200
• Schema: Simple array
• Nesting: 2 levels
• Why Excellent: Minimal, efficient branch listing. Exactly what agents need - name, SHA, protected status. No bloat, clear structure.
• Recommendation: Ideal for branch operations.

⭐⭐⭐⭐ Good Tools (Rating 4/5)

list_issues (issues)

• Tokens: 1,200
• Schema: Object with pagination
• Nesting: 4 levels
• Why Good: Rich issue data with embedded user and label objects. Good pagination support. Nesting adds some complexity but data is complete.
• Recommendation: Solid for issue querying with perPage control.

list_workflows (actions)

• Tokens: 2,400
• Schema: Object with workflows array
• Nesting: 2 levels
• Why Good: Well-structured workflow list with clear metadata (id, name, path, state). Reasonable context usage for 190 workflows.
• Recommendation: Good for discovering workflows.

list_discussions (discussions)

• Tokens: 600
• Schema: Object with pagination
• Nesting: 3 levels
• Why Good: Concise discussion list with category info. Good pagination (pageInfo with cursors). Efficient for agents to scan discussions.
• Recommendation: Use for discussion discovery.

list_commits (commits)

• Tokens: 800
• Schema: Array
• Nesting: 3 levels
• Why Good: Clean commit data with author/committer details. Good balance of detail and context efficiency.
• Recommendation: Solid for commit history.

⭐⭐⭐ Adequate Tools (Rating 3/5)

list_pull_requests (pull_requests)

• Tokens: 4,800
• Schema: Array with heavy nesting
• Nesting: 5 levels (deepest in set)
• Why Adequate: Very verbose with deep nesting. Full repo objects embedded in head/base branches. Many redundant URLs in _links object. Heavy context usage for minimal query (perPage=1).
• Issues:
  • Redundant data (full repo objects repeated)
  • Deep nesting makes parsing complex
  • _links object adds URLs already derivable from other fields
• Recommendation: Use sparingly. Consider search_pull_requests for filtering.

list_label (labels)

• Tokens: 3,200
• Schema: Object with labels array
• Nesting: 2 levels
• Why Adequate: Returns ALL labels (353 total) without pagination control. Context-heavy for what agents typically need (few specific labels). No filtering options.
• Issues:
  • No pagination despite 353 labels
  • No filtering by name pattern
  • Agents typically need 1-5 specific labels, not all 353
• Recommendation: Use get_label for specific labels if possible. Avoid for large label sets.

⭐ Poor Tools (Rating 1/5)

get_me (context)

• Tokens: 0
• Schema: Error
• Nesting: N/A
• Why Poor: Returns 403 error: "Resource not accessible by integration". Cannot provide user context to agents.
• Critical Issue: Agents cannot determine authenticated user, limiting personalization and context awareness.
• Recommendation: Fix integration permissions or document as unavailable.

Recommendations

High-Value Tools (Rating 4-5, Context-Efficient)

These tools provide excellent data with minimal context usage:

• ✅ get_file_contents - 50 tokens, rating 5/5
• ✅ list_branches - 200 tokens, rating 5/5
• ✅ list_discussions - 600 tokens, rating 4/5
• ✅ list_commits - 800 tokens, rating 4/5
• ✅ list_issues - 1,200 tokens, rating 4/5

Tools Needing Improvement

1. get_me (critical): Fix 403 error to enable user context
2. list_pull_requests: Reduce nesting, remove redundant repo objects
3. list_label: Add pagination and filtering options

Context-Heavy Tools (Use Sparingly)

• ⚠️ list_pull_requests - 4,800 tokens
• ⚠️ list_label - 3,200 tokens (all 353 labels)
• ⚠️ list_workflows - 2,400 tokens

Agent Selection Strategy

For autonomous agents optimizing context usage:

Tier 1 (Prefer): Ratings 5/5, <500 tokens

• get_file_contents, list_branches

Tier 2 (Good): Ratings 4/5, <1500 tokens

• list_discussions, list_commits, list_issues

Tier 3 (Use with caution): Ratings 3/5 or >2000 tokens

• list_pull_requests, list_label, list_workflows

Tier 4 (Avoid): Ratings 1-2/5 or errors

• get_me (fix required)

Visualizations

Response Size by Toolset

Shows average token usage across toolsets. Pull requests and labels are the most context-heavy.

Usefulness Ratings by Tool

Color-coded ratings: Green (5/5), Blue (4/5), Orange (3/5), Red (1/5). Three tools achieve perfect scores.

Token Size vs Usefulness

Ideal tools are in the bottom-left quadrant (small size, high usefulness). list_pull_requests is in the bottom-right (large size, moderate usefulness).

Nesting Depth by Tool

Lower nesting is better for agent parsing. list_pull_requests has the deepest nesting at 5 levels.

30-Day Trend Summary

Metric	Value
Data Points	10 (first run)
Analysis Date	2026-02-05
Average Token Size	1,505
Average Usefulness Rating	3.8/5
Trend Status	Baseline established

Note: This is the first analysis run. Future runs will show trends over time (token size changes, rating improvements).

References:

• Workflow Run §21709311905

AI generated by GitHub MCP Structural Analysis

• expires on Feb 12, 2026, 11:20 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-02-12T11:20:35.998Z.

Closed by Workflow