[mcp-analysis] MCP Structural Analysis - 2026-10-06

[github-actions[bot]]

Tested 10 representative GitHub MCP tool calls across 9 toolsets today. Average usefulness for agentic work was 3.2/5. The standout finding: list_code_scanning_alerts blew past the 25,000-token MCP response limit (134,672 chars / ~33.7K tokens) even with the required state=open, severity=critical,high filters applied — this toolset currently has no safe way to query a repo with many alerts. A second recurring pattern: nearly every single-item response (get_label, list_discussions, search_users, search_code) pads its payload with ~1.7KB of duplicated base64 server-icon metadata in _meta, unrelated to the actual data.

Key Metrics

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	38,132
Average Usefulness Rating	3.2/5
Best Rated Tools	github_context: 5/5
Worst Rated Tool	list_code_scanning_alerts: 1/5

Recommendations

• High-value tools (rating 4-5): github_context (zero-overhead identity), get_file_contents (with directory-listing-first guard), list_pull_requests, list_workflows (mcpscripts wrapper), search_code.
• Tools needing improvement: list_code_scanning_alerts (no pagination/head_limit escape hatch for large result sets — needs a wrapper like the list_workflows one), get_label and search_users (thin fields, high relative icon-metadata overhead).
• Context-efficient tools (low tokens, high rating): github_context (100 tok), list_workflows via mcpscripts (172 tok), list_pull_requests (70 tok).
• Context-heavy tools: list_code_scanning_alerts (33,668 tok, failed outright), get_file_contents (1,500 tok across the two-step directory-listing + file-fetch pattern).

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	38,132
Average Usefulness Rating	3.2/5
Best Rated Tool	github_context: 5/5
Worst Rated Tool	list_code_scanning_alerts: 1/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Excellent — zero-call injected identity, immediately actionable
get_file_contents	repos	⭐⭐⭐⭐	Good — large-file guard (dir listing first) avoided yesterday's README.md failure
list_pull_requests	pull_requests	⭐⭐⭐⭐	Good — compact, nested user identity, no icon bloat
list_workflows (mcpscripts)	actions	⭐⭐⭐⭐	Good — clean pagination envelope, perPage honored, no icon overhead
search_code	search	⭐⭐⭐⭐	Good — real match with highlighted fragment/indices, no rate limit this run
list_issues	issues	⭐⭐⭐	Adequate — matching issue withheld by integrity policy (safety tradeoff)
list_discussions	discussions	⭐⭐⭐	Adequate — totalCount ignores perPage scope; icon metadata ~60% of bytes
get_label	labels	⭐⭐	Limited — ~140 char payload dwarfed 5x by icon metadata
search_users	users	⭐⭐	Limited — thin fields (login/id/avatar only), needs follow-up calls
list_code_scanning_alerts	code_security	⭐	Poor — exceeded 25k token limit outright despite required filters

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
github_context	markdown	1	actor, repository, workspace, workflow-run-id	Prompt-injected, not an MCP call
get_file_contents	array+resource	2	name, path, size, type, content/resource.text	Two-step: metadata listing, then content resource
list_issues	array	2	number, title, state, labels	Empty array; 1 item filtered by integrity policy
list_pull_requests	array	3	number, title, state, user	Nested bot/user identity
list_code_scanning_alerts	error	1	—	Failed before schema could be observed
list_discussions	object	3	discussions, pageInfo, totalCount, category, user	GraphQL-style envelope
get_label	object	1	name, color, description, id	Flat, minimal payload
search_users	object	2	total_count, incomplete_results, items	Thin item fields
list_workflows	object	3	total_count, per_page, page, workflows	Clean REST pagination envelope
search_code	object	3	total_count, incomplete_results, items, text_matches	Includes highlighted fragment + match indices

Response Size Analysis

Toolset	Avg Tokens	Tools Tested
code_security	33,668	1
repos	1,500	1
discussions	700	1
labels	675	1
users	625	1
search	612	1
actions	172	1
pull_requests	70	1
workflow_context	100	1
issues	10	1

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	100	markdown	5	Zero MCP overhead, consistent day-over-day
get_file_contents	repos	1,500	array+resource	4	Directory listing (fields-limited) + LICENSE fetch; still ~1.7KB icon overhead
list_issues	issues	10	array	3	Empty visible array; 1 item integrity-filtered
list_pull_requests	pull_requests	70	array	4	Single dependabot PR, no icon bloat observed
list_code_scanning_alerts	code_security	33,668	error	1	Exceeded 25k-token limit despite required filters; no perPage/head_limit support
list_discussions	discussions	700	object	3	perPage honored for items, not totalCount; icon metadata ~60% of bytes
get_label	labels	675	object	2	~140 char payload, 5x icon-metadata overhead
search_users	users	625	object	2	Thin fields, needs follow-up calls
list_workflows (mcpscripts)	actions	172	object	4	Clean wrapper, perPage honored, no icon bloat
search_code	search	612	object	4	No rate limit this run (unlike 2026-10-05's 429); actionable fragment+indices

30-Day Trend Summary

Metric	Value
Data Points	20 (2 days: 2026-10-05, 2026-10-06)
Average Daily Tokens	23,951
Average Rating Trend	Stable (2026-10-05: 2.9/5 avg → 2026-10-06: 3.2/5 avg)

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References: §37459714970

Generated by 🔍 GitHub MCP Structural Analysis · claude · agent · 104.1 AIC · ⌖ 6.88 AIC · ⊞ 7.9K · ◷

• expires on Oct 7, 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 #66550.