[mcp-analysis] MCP Structural Analysis - 2026-09-29

[github-actions[bot]]

GitHub MCP structural analysis for 2026-09-29, testing 10 tools across every non-context toolset available to GITHUB_TOKEN. Today's data adds a new run to the 17-day, 170-datapoint rolling window (2026-08-31 → 2026-09-29). Highlights: github_context and the mcpscripts list_workflows wrapper remain best-in-class (5/5); search_code hit a 429 rate limit again (1/5); and the top-severity slice of list_code_scanning_alerts shifted composition for the first time in weeks — the opengrep "pr-action" alerts that dominated every measurement since 2026-09-15 are gone from the top 5, replaced by CodeQL Go findings.

Executive Summary

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	8,818
Average Usefulness Rating	3.3/5
Best Rated Tool	github_context / list_workflows: 5/5
Worst Rated Tool	search_code: 1/5 (429 rate limit)

Full Structural Analysis Report

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Zero-cost, immediately actionable identity block
list_workflows (mcpscripts)	actions	⭐⭐⭐⭐⭐	Clean pagination envelope, no bloat
get_file_contents	repos	⭐⭐⭐⭐	Complete file content, but icon overhead dominates small files
list_pull_requests	pull_requests	⭐⭐⭐⭐	Flat, minimal, fields-filtered, directly actionable
list_code_scanning_alerts	code_security	⭐⭐⭐	Complete but boilerplate-heavy per-alert rule.help text
list_discussions	discussions	⭐⭐⭐	Good pagination + relationships, icon overhead on small queries
get_label	labels	⭐⭐⭐	Tiny payload dwarfed by fixed icon block
search_users	users	⭐⭐⭐	Adequate for disambiguation, needs follow-up call for profile detail
list_issues	issues	⭐⭐	Integrity-filtered item hidden with no visible content
search_code	search	⭐	429 rate-limited, zero 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	Nested resource wrapper + fixed icon block
list_issues	array	2	number, title, state, labels	Empty array hides integrity-filtered item
list_pull_requests	array	1	number, title, state, html_url	Flat, fields-filtered
list_workflows	object	2	total_count, workflows[].id/name/path/state	Clean wrapper tool, explicit pagination
list_code_scanning_alerts	array	3	rule.id, rule.help, most_recent_instance.location	No icon overhead, but rule.help repeated verbatim per alert
list_discussions	object	3	discussions[].category, pageInfo, totalCount	GraphQL-style envelope w/ pagination cursors
get_label	object	1	name, color, description, id	Flat but dwarfed by icon _meta
search_users	object	2	items[].login/id/profile_url, total_count	Minimal identity fields only
search_code (failed)	error	1	error_message	429 rate limit, no fallback data

Response Size Analysis

Toolset	Avg Tokens (30d)	Tools Tested
code_security	8,032	17
repos	2,813	17
labels	751	17
discussions	708	17
users	636	17
search	635	17
actions	169	17
issues	95	17
pull_requests	92	17
workflow_context	85	17

Tool-by-Tool Analysis (Today)

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	90	markdown	5	Zero-cost identity block
get_file_contents	repos	706	object	4	CODEOWNERS unchanged since 09-25; ~75% icon overhead
list_issues	issues	70	array	2	Empty array + 1 integrity-filtered issue (#64257)
list_pull_requests	pull_requests	45	array	4	Dependabot bump PR #64242
list_workflows	actions	172	object	5	377 workflows total (up from 375)
list_code_scanning_alerts	code_security	5,100	array	3	Composition shift: opengrep alerts gone from top 5, replaced by CodeQL Go findings
list_discussions	discussions	717	object	3	totalCount 11,026 (up from 10,916 on 09-25)
get_label	labels	614	object	3	'bug' label, ~25 tokens of signal in ~614 tokens of payload
search_users	users	644	object	3	pelikhan lookup, minimal identity fields
search_code	search	660	error	1	429 rate limit, retry-after ~89s

30-Day Trend Summary

Metric	Value
Data Points	170
Days Sampled	17 (2026-08-31 → 2026-09-29)
Average Daily Tokens	~14,017
Average Rating Trend	Stable (3.5 → 3.4), today's dip driven by the search_code 429 failure

Notable Finding: Code Scanning Alert Composition Shift

Every measurement from 2026-09-15 through 2026-09-25 showed the same steady-state top-5 alerts: two recurring opengrep "pr-action" workflow findings (dockerfile-non-sha-pinned-image, github-actions-npm-install-non-deterministic/actions-uv-pip-install-non-deterministic) alongside CodeQL Go findings. Today, for the first time in this window, the opengrep findings are absent from the top 5 by severity/recency ordering — the full top 5 is now exclusively CodeQL (go/allocation-size-overflow ×4, go/bad-redirect-check ×1). This suggests either remediation of the opengrep-flagged workflow files or a re-triage that lowered their ranking; it does not necessarily mean they were resolved, since list_code_scanning_alerts was only queried with perPage=5.

Recommendations

• High-value tools (rating 4-5): github_context, list_workflows (mcpscripts wrapper), get_file_contents (small files), list_pull_requests — clean, low-overhead, actionable.
• Tools needing improvement: list_issues (integrity-filtered items disappear silently instead of surfacing a redaction placeholder in the array itself), search_code (429 reliability remains the single biggest blocker to agentic use).
• Context-efficient, high-rating tools: list_workflows (172 tokens, 5/5) and list_pull_requests (45 tokens, 4/5) — the fields-filtered, wrapper-clean pattern other tools should emulate.
• Context-heavy tools: list_code_scanning_alerts remains the single largest per-call cost (avg 8,032 tokens/call over 30 days) due to verbatim-repeated rule.help markdown; deduplicating help text by rule ID rather than inlining it per-alert-instance would cut this by roughly 4-5x for repeat-rule alert sets.

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Generated by 🔍 GitHub MCP Structural Analysis · claude · agent · 175.3 AIC · ⌖ 7.31 AIC · ⊞ 7.7K · ◷

• expires on Sep 30, 2026, 4:05 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 #64484.