[mcp-analysis] MCP Structural Analysis - 2026-10-13

[github-actions[bot]]

Overview

Tested 10 representative GitHub MCP tools (one per toolset) on 2026-10-13, plus the injected <github-context> workflow-identity block. Average usefulness rating was 3.30/5. Two tools (list_issues, list_code_scanning_alerts) surfaced a notable governance finding: real data existed server-side but was silently filtered by integrity/secrecy policy, leaving an empty-looking response — a false-negative risk for automated triage.

Key Metrics

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	5,529
Average Usefulness Rating	3.30/5
Best Rated Tools	github_context, list_pull_requests, list_workflows — 5/5
Worst Rated Tools	list_issues, list_code_scanning_alerts, search_users — 2/5

Full Structural Analysis Report

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Excellent — zero-cost, directly actionable identity metadata
list_pull_requests	pull_requests	⭐⭐⭐⭐⭐	Excellent — fields selection stripped all icon/_meta bloat
list_workflows (mcpscripts)	actions	⭐⭐⭐⭐⭐	Excellent — clean JSON, explicit pagination, no bloat
get_file_contents	repos	⭐⭐⭐	Adequate — works for small files, but duplicates body 2x
list_discussions	discussions	⭐⭐⭐	Adequate — good cursor pagination, heavy fixed overhead
get_label	labels	⭐⭐⭐	Adequate — clean payload, icon overhead dominates
search_code	search	⭐⭐⭐	Adequate — fields filter avoided rate limit, lost snippet context
list_issues	issues	⭐⭐	Limited — only result was policy-filtered, empty array
list_code_scanning_alerts	code_security	⭐⭐	Limited — only alert was policy-filtered, empty array
search_users	users	⭐⭐	Limited — no dedicated get_user tool; worst payload-to-overhead ratio

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, structuredContent, resource.text	Body duplicated 2x
list_issues	array	1	(filtered)	Integrity-policy filter hid sole result
list_pull_requests	array	2	number, title, state, head.ref, base.ref	Clean with fields filter
list_code_scanning_alerts	array	1	(filtered)	Secrecy-policy filter hid sole result
list_discussions	object	3	discussions[], pageInfo, totalCount	Cursor pagination well-formed
get_label	object	1	name, color, description	Flat, trivial payload
list_workflows	object	2	total_count, per_page, workflows[]	mcpscripts wrapper, needs explicit owner/repo
search_users	object	2	items[].login, items[].id, total_count	Substituted for missing get_user
search_code	object	2	items[].name, items[].path, total_count	fields filter dropped text_matches

Response Size Analysis

Toolset	Avg Tokens (30-day)	Tools Tested
repos	3,240	2
discussions	980	2
users	929	2
labels	893	2
search	546	3
issues	410	2
pull_requests	410	2
actions	158	2
code_security	85	2
workflow_context	75	2

Tool-by-Tool Analysis (2026-10-13)

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	60	markdown	5	Mostly #false placeholders this run (non-issue/PR trigger)
get_file_contents	repos	1,380	object	3	Used LICENSE (1060 bytes) to dodge yesterday's README overflow
list_issues	issues	70	array	2	Sole issue (#68395) integrity-filtered
list_pull_requests	pull_requests	120	array	5	fields param eliminated all wrapper overhead
list_code_scanning_alerts	code_security	80	array	2	Sole alert secrecy-filtered
list_discussions	discussions	1,010	object	3	totalCount=11,382, good cursor pagination
get_label	labels	885	object	3	Icon/_meta overhead ~97% of payload
list_workflows	actions	140	object	5	Best token-to-value ratio measured
search_users	users	907	object	2	No dedicated get_user tool exists
search_code	search	877	object	3	Fields filter avoided prior rate-limit failure

30-Day Trend Summary

Metric	Value
Data Points	21
Average Daily Tokens	7,997
Average Rating Trend	Improving — fields parameter usage today cut list_pull_requests overhead to near zero vs. unfiltered calls yesterday

Recommendations

• High-value tools (rating 4–5): github_context (no-op, use first), list_pull_requests and list_workflows when called with explicit fields/wrapper — confirms fields selection is the single biggest token-efficiency lever on this server.
• Tools needing improvement: list_issues and list_code_scanning_alerts can return a visually "empty" result while hiding policy-filtered data — agents should check for filter notices, not just treat [] as "nothing found." search_users has no lightweight single-user equivalent.
• Context-efficient tools: list_workflows (mcpscripts wrapper, 140 tokens, rating 5) and list_pull_requests with fields (120 tokens, rating 5).
• Context-heavy tools: get_file_contents (repos) and list_discussions carry a fixed ~3.4KB icon/_meta overhead per call regardless of payload size — batch related reads where possible to amortize this cost.

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Generated by 🔍 GitHub MCP Structural Analysis · claude · agent · 110.2 AIC · ⊞ 8K · ◷

• expires on Oct 14, 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 #68689.