[mcp-analysis] MCP Structural Analysis - 2026-10-01

[github-actions[bot]]

MCP Structural Analysis — 2026-10-01

Tested 10 representative GitHub MCP tools across all available toolsets. Today's total payload was 6,867 tokens across 10 calls, with an average usefulness rating of 3.3/5. The workflow-injected <github-context> block and the mcpscripts list_workflows wrapper remain the best-rated sources (5/5, near-zero overhead); search_code scored worst (1/5) due to a 429 rate-limit failure that still cost ~665 tokens for an empty error envelope.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	6,867
Average Usefulness Rating	3.3/5
Best Rated Tool	github_context / list_workflows: 5/5
Worst Rated Tool	search_code: 1/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Zero-cost injected identity block, always actionable
list_workflows (mcpscripts)	actions	⭐⭐⭐⭐⭐	Clean pagination envelope, no icon/_meta bloat
get_file_contents	repos	⭐⭐⭐⭐	Correct content, but icon _meta block dominates payload
list_pull_requests	pull_requests	⭐⭐⭐⭐	Flat, minimal, directly actionable
list_code_scanning_alerts	code_security	⭐⭐⭐	Usable but rule.help boilerplate repeated per alert
list_discussions	discussions	⭐⭐⭐	Good pagination metadata, icon overhead dominates small payload
search_users	users	⭐⭐⭐	Minimal identity fields, needs follow-up call for profile detail
get_label	labels	⭐⭐⭐	Tiny real payload drowned by fixed icon overhead
list_issues	issues	⭐⭐	Integrity-filtered item hidden with no visible indication of redaction
search_code	search	⭐	429 rate limit, zero usable data, still paid full overhead

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
github_context	markdown	1	actor, repository, workspace, workflow-run-id	Prompt-injected, no MCP call
get_file_contents	object	3	content.resource.text, _meta.serverInfo	Fixed ~2.5KB base64 icon block overhead
list_issues	array	2	number, title, state, labels	Redacted item silently dropped from array
list_pull_requests	array	1	number, title, state, html_url	Flat, no nesting
list_workflows (wrapper)	object	2	total_count, per_page, workflows[]	Raw stdout JSON, no envelope overhead
list_code_scanning_alerts	array	3	rule.id, rule.full_description, rule.help, most_recent_instance.location	Verbatim rule.help repeated per alert of same rule
list_discussions	object	3	discussions[].category.name, pageInfo.hasNextPage, totalCount	Nested category/user relationships
get_label	object	1	name, color, description, id	Simple flat object
search_users	object	2	items[].login, items[].profile_url, total_count	No name/bio in minimal mode
search_code	error	1	error_message	429 rate-limit envelope

Response Size Analysis (today)

Toolset	Tokens	Tools Tested
code_security	3,150	1
discussions	720	1
repos	700	1
search	665	1
users	644	1
labels	614	1
actions	172	1
issues	72	1
pull_requests	45	1
workflow_context	85	1

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	85	markdown	5	Zero-cost, consistent 22/22 days
list_workflows	actions	172	object	5	mcpscripts wrapper, best-in-class, 22/22 days
get_file_contents	repos	700	object	4	CODEOWNERS targeted after size pre-check; icon overhead ~75%
list_pull_requests	pull_requests	45	array	4	Spec-extractor PR #64740
list_code_scanning_alerts	code_security	3150	array	3	Same 3 alerts (#943-945) as 2026-09-30, no remediation
list_discussions	discussions	720	object	3	totalCount 11079 (up from 11052)
search_users	users	644	object	3	pelikhan lookup, minimal fields
get_label	labels	614	object	3	"bug" label, tiny payload + fixed overhead
list_issues	issues	72	array	2	Issue #64754 redacted by integrity policy
search_code	search	665	error	1	429 rate limit, retry ~7.3s

30-Day Trend Summary

Metric	Value
Data Points	180
Days Measured	18
Average Daily Tokens	14,419
Average Rating Trend	Stable (first half 3.47 → second half 3.37)

Highest average-token toolset over the window: code_security (~7,935 tokens/call), driven almost entirely by verbatim rule.help markdown repeated per alert of the same rule.id. Lowest: workflow_context (~85 tokens/call), the injected context block.

Recommendations

• High-value tools (rating 4-5): github_context (zero-cost identity), list_workflows via the mcpscripts wrapper (clean pagination, no overhead), list_pull_requests (flat/actionable with field selection), get_file_contents on small, pre-size-checked files.
• Tools needing improvement: list_code_scanning_alerts would benefit from deduplicating rule.help across alerts sharing a rule.id (could roughly halve payload for repeated-rule result sets); list_issues should surface a visible marker when an item is dropped by integrity filtering rather than silently returning an empty array; search_code needs backoff/retry guidance given persistent 429s.
• Context-efficient tools (low tokens, high rating): github_context, list_workflows (wrapper), list_pull_requests.
• Context-heavy tools: list_code_scanning_alerts (3,150 tokens for only 3 alerts, due to repeated help text), get_file_contents and other non-wrapper calls (fixed ~2.5KB base64 icon _meta block adds overhead to every raw GitHub MCP response regardless of payload size).

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Generated by 🔍 GitHub MCP Structural Analysis · claude · agent · 130.5 AIC · ⌖ 9.69 AIC · ⊞ 8.1K · ◷

• expires on Oct 2, 2026, 4:03 AM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by GitHub MCP Structural Analysis.

A newer discussion is available at Discussion #65012.