[mcp-analysis] MCP Structural Analysis - 2026-02-12

[github-actions[bot]]

Analyzed 15 GitHub MCP tools for response size and structural usefulness in agentic workflows. Average rating: 4.47/5. Best performers: get_file_contents (25 tokens, 5/5), get_label (50 tokens, 5/5). Context-heavy tools: list_releases (7,551 tokens), list_pull_requests (5,575 tokens), list_workflows (3,885 tokens with broken pagination).

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	15
Total Tokens (Today)	27,836
Average Usefulness Rating	4.47/5 ⭐⭐⭐⭐½
Best Rated Tool	get_file_contents: 5/5 ⭐⭐⭐⭐⭐
Most Efficient	get_file_contents (25 tokens, 5/5)
Largest Response	list_releases (7,551 tokens)

Usefulness Ratings for Agentic Work

Rating Scale:

• ⭐⭐⭐⭐⭐ (5): Excellent - Complete, actionable, immediately usable
• ⭐⭐⭐⭐ (4): Good - Mostly complete, minor gaps or redundancy
• ⭐⭐⭐ (3): Adequate - Usable but context-heavy or requires supplementary calls
• ⭐⭐ (2): Limited - Missing key data or hard to parse
• ⭐ (1): Poor - Minimal value or unavailable in CI context

Tool	Toolset	Rating	Tokens	Assessment
get_file_contents	repos	⭐⭐⭐⭐⭐	25	Perfect: minimal overhead, clean text response
get_label	labels	⭐⭐⭐⭐⭐	50	Minimal response with complete label metadata
list_branches	repos	⭐⭐⭐⭐⭐	100	Clean array with essential data only
list_tags	repos	⭐⭐⭐⭐⭐	100	Efficient tag list with commit SHA and URLs
get_commit	repos	⭐⭐⭐⭐⭐	200	Clean metadata (include_diff=false option)
list_discussions	discussions	⭐⭐⭐⭐⭐	275	GraphQL with cursor pagination, well-organized
list_commits	repos	⭐⭐⭐⭐⭐	325	Good balance of commit details
search_repositories	search	⭐⭐⭐⭐⭐	350	Excellent discovery with rich metadata
list_issues	issues	⭐⭐⭐⭐⭐	1,650	Rich GraphQL response, complete and actionable
issue_read	issues	⭐⭐⭐⭐⭐	1,950	Complete single issue with full body and reactions
search_issues	search	⭐⭐⭐⭐	2,600	Comprehensive but some redundancy in user fields
search_pull_requests	search	⭐⭐⭐⭐	3,200	Good discovery, some redundant user data
list_workflows	actions	⭐⭐⭐	3,885	Pagination broken: returns all 192 workflows
list_pull_requests	pull_requests	⭐⭐⭐	5,575	Heavy redundancy: full repo objects duplicated
list_releases	releases	⭐⭐⭐	7,551	Context-heavy: full release notes and assets

Schema Analysis

Tool	Type	Depth	Key Fields	Schema Notes
get_file_contents	string	1	content	Flat string, no nesting
get_label	object	1	id, name, color, description	Simple object
list_branches	array	2	name, sha, protected	Clean array
list_tags	array	2	name, commit.sha, urls	Minimal nesting
get_commit	object	3	sha, message, author	Reasonable nesting
list_discussions	object	3	discussions[], pageInfo	GraphQL pagination
list_commits	array	3	sha, commit, author	Consistent structure
search_repositories	object	3	total_count, items[]	Standard search format
list_issues	object	4	issues[], pageInfo, totalCount	GraphQL with rich details
issue_read	object	3	id, title, body, labels, reactions	Well-organized single entity
search_issues	object	4	total_count, items[], bodies	Full issue bodies included
search_pull_requests	object	4	total_count, items[], bodies	Comprehensive PR data
list_workflows	oversized	2	workflows[]	Ignores perPage parameter
list_pull_requests	oversized	5	head.repo, base.repo	Deep nesting with duplication
list_releases	oversized	4	body, assets[], author	Full release notes included

Response Size Analysis

Toolset	Avg Tokens	Tools Tested	Assessment
releases	7,551	1	Context-heavy: full release bodies
pull_requests	5,575	1	Heavy redundancy in repo objects
actions	3,885	1	Broken pagination, returns all workflows
search	2,037	3	Varies: repos efficient, issues/PRs verbose
issues	1,800	2	GraphQL responses, good detail balance
repos	155	5	Highly efficient (excl. tree endpoint)
discussions	275	1	Clean GraphQL response
labels	50	1	Minimal and efficient

Tool-by-Tool Analysis

🏆 Top Performers (Rating 5/5, Low Context)

get_file_contents (25 tokens)

• Returns plain text file contents
• Minimal overhead, no metadata bloat
• Perfect for agents needing code/config files

get_label (50 tokens)

• Simple object: id, name, color, description
• No unnecessary fields
• Ideal for label operations

list_branches (100 tokens)

• Array with name, SHA, protected status
• Essential data only
• Efficient for branch discovery

list_tags (100 tokens)

• Clean tag list with commit SHAs
• Includes download URLs
• Perfect for release tracking

get_commit (200 tokens)

• Option to exclude diff saves massive context
• Good author/committer details
• Excellent for commit metadata

⚙️ Well-Designed Tools (Rating 5/5)

list_discussions (275 tokens)

• GraphQL with cursor pagination
• Includes category, user, timestamps
• Well-structured for agents

list_commits (325 tokens)

• Clean commit history
• Balanced detail level
• Good for analysis tasks

search_repositories (350 tokens)

• Rich metadata: stars, topics, language
• All needed discovery data
• Immediately actionable

list_issues (1,650 tokens)

• GraphQL with full issue bodies
• Includes labels, user info, pagination
• Complete and actionable

issue_read (1,950 tokens)

• Full single issue with reactions
• Clean structure, good detail balance
• Perfect for issue operations

👍 Good Tools (Rating 4/5)

search_issues (2,600 tokens)

• Comprehensive with full bodies
• Some redundancy in user objects (multiple URL fields)
• Still actionable for finding issues

search_pull_requests (3,200 tokens)

• Good PR discovery
• Includes labels, assignees, reactions
• Some redundant user data but usable

⚠️ Needs Improvement (Rating 3/5)

list_workflows (3,885 tokens)

• Critical issue: Pagination broken
• Ignores perPage=1, returns all 192 workflows
• Should return 1 workflow, not entire list
• Data is useful but inefficient

list_pull_requests (5,575 tokens)

• Critical issue: Heavy redundancy
• Full repo objects for both head and base
• Duplicates ALL repo fields (urls, permissions, etc.)
• Could be optimized to just repo IDs/names

list_releases (7,551 tokens)

• Context-heavy: full release body with markdown
• All assets with complete metadata
• Nested user objects
• Data is useful but expensive

30-Day Trend Summary

Metric	Value
Data Points	47 (across 4 analysis days)
Unique Tools Tested	23
Average Daily Tokens	~7,000-28,000 (varies by tools tested)
Rating Trend	Stable (4.3-4.5 average)
Days with Data	4 (Feb 6, 10, 11, 12)

Key Findings from Trends:

• Most tools maintain consistent response sizes
• list_pull_requests tokens increased (2,800 → 5,575)
• list_workflows size varies but pagination always broken
• GraphQL-based tools (issues, discussions) remain efficient
• Security tools (list_code_scanning_alerts: 23,800 tokens) are extremely context-heavy

Recommendations

For Agent Developers

High-value, context-efficient tools:

• get_file_contents: Best for reading code/configs (25 tokens)
• get_label: Efficient label operations (50 tokens)
• list_branches: Quick branch discovery (100 tokens)
• list_tags: Version tracking (100 tokens)
• get_commit (with include_diff=false): Metadata only (200 tokens)

High-value tools (moderate context):

• list_discussions: Well-structured (275 tokens)
• list_commits: Good commit history (325 tokens)
• search_repositories: Rich discovery (350 tokens)
• list_issues: Complete issue data (1,650 tokens)
• issue_read: Single issue details (1,950 tokens)

Use with caution (context-heavy):

• search_issues: Consider filtering to reduce body size (2,600 tokens)
• search_pull_requests: Similar concern (3,200 tokens)
• list_workflows: Avoid if possible - pagination broken (3,885 tokens)
• list_pull_requests: Heavy redundancy, consider alternatives (5,575 tokens)
• list_releases: Very expensive, only use when necessary (7,551 tokens)

Avoid in agentic workflows:

• list_code_scanning_alerts: Massive payload (23,800 tokens)
• list_workflow_runs: Extreme context cost (90,096 tokens)
• get_me, list_notifications, projects_list: Unavailable in CI (403/404 errors)

For MCP Server Developers

Pagination Issues:

• list_workflows: Does not respect perPage parameter
• list_workflow_runs: Ignores perPage, returns massive dataset
• Recommendation: Fix pagination to honor perPage limits

Redundancy Issues:

• list_pull_requests: Duplicates full repo objects for head and base
• Recommendation: Return repo IDs/names only, or add include_full_repo parameter

Context Optimization:

• list_releases: Consider adding include_body parameter to exclude release notes
• search_issues/search_pull_requests: Add parameter to exclude full bodies
• list_code_scanning_alerts: Add aggressive filtering or summary mode

Error Handling:

• Several tools return 403/404 in CI contexts
• Recommendation: Add error messages explaining availability restrictions

Visualizations

Response Size by Toolset

Usefulness Ratings by Toolset

30-Day Token Usage Trend

Token Size vs Usefulness

References:

• §21944374158

AI generated by GitHub MCP Structural Analysis

• expires on Feb 19, 2026, 11:22 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-02-19T11:22:08.677Z.

Closed by Workflow