[mcp-analysis] GitHub MCP Structural Analysis - February 13, 2026

[github-actions[bot]]

Analyzed 19 GitHub MCP tools across 10 toolsets to evaluate response sizes and structural usefulness for agentic workflows. Average usefulness rating: 4.05/5 with 10 tools rated excellent (5/5). Total token usage today: 29,488 tokens. Key finding: Most efficient tools are get_file_contents (25 tokens, 5/5), get_label (50 tokens, 5/5), and list_branches (100 tokens, 5/5). Context-heavy tools include list_releases (7,058 tokens), list_pull_requests (5,125 tokens), and list_workflows (3,885 tokens) due to broken pagination or nested redundancy.

View Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	19
Toolsets Covered	10 (repos, issues, pull_requests, actions, discussions, search, releases, labels, notifications, context)
Total Tokens (Today)	29,488
Average Tokens per Tool	1,552
Average Usefulness Rating	4.05 / 5
Best Rated Tool	get_file_contents: 5/5 ⭐⭐⭐⭐⭐ (25 tokens)
Worst Rated Tools	get_me, list_notifications: 1/5 ⭐ (403 errors)
Context-Efficient Champions	get_file_contents (25 tokens), get_label (50 tokens), list_branches (100 tokens)
Context-Heavy Tools	list_releases (7,058 tokens), list_pull_requests (5,125 tokens), list_workflows (3,885 tokens)

Usefulness Ratings for Agentic Work

Tools rated by their structural usefulness for autonomous agent workflows (1-5 scale).

⭐⭐⭐⭐⭐ Excellent (Rating 5/5) - 10 Tools

These tools provide complete, actionable data with clean structures. Immediately usable by agents.

Tool	Toolset	Tokens	Assessment
get_file_contents	repos	25	Clean text response. Returns file contents directly. Perfect efficiency.
get_label	labels	50	Minimal response with just label metadata. No unnecessary fields.
list_branches	repos	100	Essential data: name, SHA, protection status. Perfect balance.
list_tags	repos	100	Clean tag list with commit SHA and download URLs.
get_commit (no diff)	repos	200	Efficient commit metadata without diff. Good author details.
list_discussions	discussions	275	GraphQL with cursor pagination. Clean category/user/timestamps.
list_commits	repos	325	Clean commit history with message, author, committer.
search_repositories	search	350	Rich metadata: stars, topics, language, dates. Excellent discovery.
issue_read	issues	950	Complete issue details: body, reactions, assignees, labels.
list_issues	issues	1,650	GraphQL with cursor pagination. Rich issue details, well-structured.

⭐⭐⭐⭐ Good (Rating 4/5) - 5 Tools

Good data with minor gaps or redundancy. Mostly actionable but room for improvement.

Tool	Toolset	Tokens	Assessment
search_code	search	950	Good discovery tool. Full repo context per result. Some redundancy.
search_issues	search	2,600	Comprehensive results with full bodies. Extensive user info with URL bloat.
get_repository_tree	repos	2,645	Clean array structure (46 items). Good for directory exploration.
search_pull_requests	search	3,200	Comprehensive PR search. Full bodies, labels, reactions. Some redundancy.
get_repository_tree	repos	2,645	Clean directory structure. Non-recursive mode is efficient.

⭐⭐⭐ Adequate (Rating 3/5) - 3 Tools

Usable but requires supplementary calls or has significant inefficiencies.

Tool	Toolset	Tokens	Assessment
list_workflows	actions	3,885	Pagination broken: Returns all 193 workflows despite perPage=1. Context-heavy.
list_pull_requests	pull_requests	5,125	Oversized payload: Full nested repo objects for head and base. Heavy redundancy.
list_releases	releases	7,058	Massive payload: Full release body, all assets, nested user objects. Context-heavy.

⭐ Poor (Rating 1/5) - 2 Tools

Unavailable or broken in GitHub Actions context. Not usable for agents in CI.

Tool	Toolset	Tokens	Assessment
get_me	context	0	403 error: Not accessible by integration. Unavailable in Actions.
list_notifications	notifications	0	403 error: Not accessible by integration. Unavailable in Actions.

Schema Analysis

Response structures and organization patterns identified:

Tool	Type	Depth	Key Fields	Schema Notes
get_file_contents	string	1	content	Direct text response. Minimal overhead.
get_label	object	1	id, name, color, description	Flat object. No nesting.
list_branches	array	2	name, sha, protected	Simple array of objects.
list_issues	object	4	issues[], pageInfo, totalCount	GraphQL structure with cursor pagination.
list_pull_requests	oversized	5	id, number, head.repo, base.repo	Deep nesting with full repo duplication.
list_workflows	oversized	2	total_count, workflows[]	Flat workflows array but ignores pagination.
search_repositories	object	3	total_count, items[].id, items[].name	Standard search result structure.
get_repository_tree	oversized	2	sha, tree[].path, tree[].type	Clean array but large (46 items).

Schema Patterns Observed:

• GraphQL tools (list_issues, list_discussions): Excellent structure with cursor pagination and clean response shapes
• REST search tools: Good metadata but include extensive user info with many URL fields (some redundancy)
• List tools with nesting: Often include full nested objects (e.g., repo objects in PRs) causing context bloat
• Actions tools: Pagination appears broken, returning full datasets regardless of perPage parameter

Response Size Analysis

Token usage by toolset (average per tool):

Toolset	Avg Tokens	Tools Tested	Size Category
releases	7,058	1	🔴 Very Heavy
pull_requests	5,125	1	🔴 Very Heavy
actions	3,885	1	🔴 Heavy
search	1,675	4	🟡 Moderate
issues	1,183	2	🟡 Moderate
repos	617	6	🟢 Light
discussions	275	1	🟢 Very Light
labels	50	1	🟢 Very Light
context	0	1	⚪ Error
notifications	0	1	⚪ Error

Tool-by-Tool Detailed Analysis

Complete reference table for all 19 tools analyzed today:

Tool	Toolset	Tokens	Schema	Depth	Rating	Key Observations
get_file_contents	repos	25	string	1	⭐⭐⭐⭐⭐	Perfect efficiency. Returns raw content.
get_label	labels	50	object	1	⭐⭐⭐⭐⭐	Minimal metadata. No bloat.
list_branches	repos	100	array	2	⭐⭐⭐⭐⭐	Essential data only.
list_tags	repos	100	array	2	⭐⭐⭐⭐⭐	Clean tag list with SHA.
get_commit	repos	200	object	3	⭐⭐⭐⭐⭐	Without diff: efficient metadata.
list_discussions	discussions	275	object	3	⭐⭐⭐⭐⭐	GraphQL pagination. Clean structure.
list_commits	repos	325	array	3	⭐⭐⭐⭐⭐	Good detail balance.
search_repositories	search	350	object	3	⭐⭐⭐⭐⭐	Rich discovery metadata.
search_code	search	950	object	4	⭐⭐⭐⭐	Good but repo context adds size.
issue_read	issues	950	object	3	⭐⭐⭐⭐⭐	Complete issue details.
list_issues	issues	1,650	object	4	⭐⭐⭐⭐⭐	GraphQL. Rich and well-structured.
search_issues	search	2,600	object	4	⭐⭐⭐⭐	Full bodies. User info verbose.
get_repository_tree	repos	2,645	oversized	2	⭐⭐⭐⭐	46 items. Clean structure.
search_pull_requests	search	3,200	object	4	⭐⭐⭐⭐	Comprehensive. Some redundancy.
list_workflows	actions	3,885	oversized	2	⭐⭐⭐	Pagination broken. Returns all 193.
list_pull_requests	pull_requests	5,125	oversized	5	⭐⭐⭐	Heavy redundancy. Nested repos.
list_releases	releases	7,058	oversized	4	⭐⭐⭐	Massive payload. Full notes/assets.
get_me	context	0	error	0	⭐	403 error. Unavailable in Actions.
list_notifications	notifications	0	error	0	⭐	403 error. Unavailable in Actions.

30-Day Trend Summary

Data collected from February 6 - February 13, 2026 (8 days, 66 data points):

Metric	Value
Total Data Points	66 entries
Date Range	Feb 6 - Feb 13, 2026
Unique Tools Analyzed	22
Unique Toolsets	12
Average Daily Tokens	~28,000 - 30,000
Rating Trend	Stable at 4.0-4.1 average
Most Tested Tool	get_file_contents (8 times)
Least Tested Toolsets	context, notifications, projects (errors)

Trends Observed:

• Consistent ratings: Most tools maintain stable usefulness ratings across multiple days
• Size variation: Some tools show size variation (e.g., list_releases: 7,058-7,551 tokens) due to content changes
• Persistent issues: Pagination problems in list_workflows and list_pull_requests remain across all test days
• Error consistency: Authentication-restricted tools (get_me, list_notifications) consistently fail in Actions context

Recommendations

Based on structural analysis and usefulness ratings:

✅ High-Value Tools (Rating 4-5, Recommended for Agents)

Context-Efficient Champions (< 500 tokens, rating 5/5):

• get_file_contents (25 tokens) - File reading
• get_label (50 tokens) - Label operations
• list_branches (100 tokens) - Branch listing
• list_tags (100 tokens) - Tag listing
• get_commit with include_diff=false (200 tokens) - Commit metadata
• list_discussions (275 tokens) - Discussion listing
• list_commits (325 tokens) - Commit history
• search_repositories (350 tokens) - Repo discovery

Feature-Rich High-Value Tools (> 500 tokens, rating 5/5):

• issue_read (950 tokens) - Single issue details
• list_issues (1,650 tokens) - Issue listing with GraphQL

Good Tools Worth Using (rating 4/5):

• search_code (950 tokens) - Code discovery
• search_issues (2,600 tokens) - Issue search across repos
• get_repository_tree (2,645 tokens) - Directory exploration
• search_pull_requests (3,200 tokens) - PR discovery

⚠️ Tools Needing Improvement (Rating 3/5)

Pagination Issues:

• list_workflows - Action Required: Fix pagination. Currently returns all 193 workflows regardless of perPage parameter
• list_pull_requests - Action Required: Reduce nested repo duplication. Consider lightweight response option
• list_releases - Action Required: Add option to exclude full release body or assets for listing operations

Recommendations for Improvement:

1. Add minimal=true parameter to return only essential fields
2. Fix pagination in Actions toolset (list_workflows, list_workflow_runs)
3. Reduce nested object duplication in PR responses
4. Consider separate endpoints for "list metadata" vs "get full details"

❌ Unavailable Tools (Rating 1/5)

Authentication-Restricted:

• get_me - Not accessible by GitHub Actions integration token
• list_notifications - Not accessible by GitHub Actions integration token
• projects_list - 404 errors in many contexts

Recommendations:

• Document which tools require specific authentication scopes
• Provide clear error messages indicating why tools are unavailable
• Consider alternative approaches for agents running in CI contexts

🎯 Context Optimization Strategy

For Context-Conscious Agents:

1. Prefer GraphQL-based tools (list_issues, list_discussions) for clean structures and good pagination
2. Use include_diff=false when calling get_commit to save tokens
3. Avoid list_releases unless full release notes are needed - consider list_tags + get_release_by_tag instead
4. Use search tools sparingly as they include verbose user objects (many URL fields)
5. Test pagination parameters before relying on them (Actions tools ignore them)

Context Budget Guidelines:

• < 100 tokens: Excellent efficiency (get_file_contents, get_label, list_branches, list_tags)
• 100-500 tokens: Good efficiency (get_commit, list_discussions, list_commits, search_repositories)
• 500-2,000 tokens: Moderate efficiency (issue_read, list_issues, search_issues)
• 2,000-5,000 tokens: Heavy usage (get_repository_tree, search_pull_requests, list_workflows, list_pull_requests)
• > 5,000 tokens: Very heavy (list_releases) - use cautiously

Visualizations

Response Size by Toolset

Average token usage per toolset. Shows releases, pull_requests, and actions as the heaviest toolsets.

Usefulness Ratings by Toolset

Average usefulness ratings (1-5 scale). Green bars (≥4) indicate good-to-excellent tools for agents. Orange/red bars indicate tools needing improvement.

Daily Token Usage Trend

Total token usage over the 8-day analysis period. Shows stable daily patterns around 28K-30K tokens per analysis run.

Token Size vs Usefulness Rating

Scatter plot showing the relationship between token size and usefulness. Ideal tools are in the lower-right (small size, high rating). Upper-left quadrant shows context-heavy, low-value tools to avoid.

Rating Distribution

Distribution of usefulness ratings. Shows most tools (10) are rated excellent (5 stars), with good balance across rating categories.

Analysis Methodology: Each tool tested with minimal parameters (perPage=1 where applicable) to measure baseline response size and structure. Token counts estimated at 1 token ≈ 4 characters. Usefulness ratings based on completeness, actionability, clarity, efficiency, and relationship inclusion. Data persisted in cache memory with 30-day rolling window.

References:

• §21984726245

AI generated by GitHub MCP Structural Analysis

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

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-02-20T11:20:24.372Z.

Closed by Workflow