[mcp-analysis] MCP Structural Analysis - 2026-08-12

[github-actions[bot]]

Analyzed 10 representative GitHub MCP tool calls across 9 toolsets (plus the injected workflow-context block). Average usefulness: 3.3/5. Best: github_context (5/5, zero MCP overhead). Worst: get_teams (1/5 — blocked by a permission gate, returned no data at all). Two structural issues stood out: every MCP response repeats ~2-4 KB of identical base64 server-info icons in _meta, and the list_workflows wrapper's per_page: 1 was silently ignored (30 full workflow objects came back instead of 1).

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	13,485
Average Usefulness Rating	3.3/5
Best Rated Tool	github_context: 5/5
Worst Rated Tool	get_teams: 1/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Excellent — free, immediately actionable identity data
list_pull_requests	pull_requests	⭐⭐⭐⭐	Good — clean, field-filtered, minimal
list_code_scanning_alerts	code_security	⭐⭐⭐⭐	Good — correctly scoped, efficient even when empty
list_discussions	discussions	⭐⭐⭐⭐	Good — category, author, cursor pagination all present
search_code	search	⭐⭐⭐⭐	Good — text-match fragments with highlight indices
get_file_contents	repos	⭐⭐⭐	Adequate — content is clean but padded with icon + unrelated README contributor list
list_issues	issues	⭐⭐⭐	Adequate — minimal, but one match was silently redacted by an integrity policy filter
list_label	labels	⭐⭐⭐	Adequate — no per-page control, dumped all 868 labels in one call
list_workflows	actions	⭐⭐	Limited — pagination param ignored, returned 30x more than requested
get_teams	context	⭐	Poor — blocked by permission gate, zero usable data

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
github_context	markdown	1	actor, repository, workspace, workflow-run-id	Flat bullet list, no MCP call needed
get_file_contents	object	4	content, resource.text, resource.uri, _meta.serverInfo	Nested resource wrapper + icon metadata
list_issues	array	2	number, title, state	Field-filtered; integrity-redaction notices interleaved with real items
list_pull_requests	array	2	number, title, state	Field-filtered, flat
list_workflows	object	3	total_count, workflows[].id/name/path/state/html_url	Rich per-workflow metadata, ignored pagination
list_code_scanning_alerts	array	1	(none — empty)	Trivially flat when no alerts match
list_discussions	object	3	discussions[].category/title/user, pageInfo, totalCount	Cursor pagination + relationship (category, author) included
list_label	object	2	labels[].name/color/description, totalCount	Flat per-label objects, count-only pagination signal
search_code	object	3	total_count, items[].name/path/text_matches	Match fragments with character indices — directly actionable

Response Size Analysis

Toolset	Avg Tokens	Tools Tested
repos	4,500	1
actions	3,300	1
labels	3,250	1
search	1,150	1
discussions	1,100	1
workflow_context	90	1
issues	60	1
pull_requests	25	1
code_security	10	1
context	0	1 (blocked)

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	90	markdown	5	Injected, no call overhead
get_teams	context	0	error	1	Permission not granted this run
get_file_contents	repos	4,500	object	3	Icon + contributor-list bloat
list_issues	issues	60	array	3	Integrity-filtered item, not a clean miss
list_pull_requests	pull_requests	25	array	4	Minimal and clean
list_workflows	actions	3,300	object	2	Pagination param ignored
list_code_scanning_alerts	code_security	10	array	4	Efficient, correctly scoped
list_discussions	discussions	1,100	object	4	Cursor pagination present
list_label	labels	3,250	object	3	No pagination control, 868 labels
search_code	search	1,150	object	4	Highlighted match fragments

30-Day Trend Summary

Metric	Value
Data Points	10 (first day of tracking)
Average Daily Tokens	13,485
Average Rating Trend	baseline (no prior data to compare)

This is the first entry in the rolling 30-day dataset at /tmp/gh-aw/cache-memory/mcp_analysis.jsonl (cache-memory persists it across future runs).

Recommendations

• High-value tools (rating 4-5): github_context, list_pull_requests, list_code_scanning_alerts, list_discussions, search_code — prefer these when their data suffices, especially the free context block over any team/user lookup.
• Tools needing improvement: get_teams (permission-gated, unusable without prior approval), list_workflows via the mcp-scripts wrapper (pagination not respected — verify per_page vs perPage behavior before relying on it for large workflow lists), list_label (no pagination primitive at all despite an 868-label repo).
• Context-efficient tools (low tokens, high rating): github_context (90 tokens, 5/5), list_pull_requests (25 tokens, 4/5), list_code_scanning_alerts (10 tokens, 4/5).
• Context-heavy tools: get_file_contents (4,500 tokens, driven by embedded icon metadata and an unrelated README contributor list, not the file content itself), list_workflows (3,300 tokens due to ignored pagination), list_label (3,250 tokens, all labels returned every call).
• Structural note: nearly every MCP response repeats a multi-KB base64 server icon block in _meta.serverInfo. Since this is identical across calls, a client-side or gateway-side strip/cache of _meta would meaningfully cut real-world token overhead across all toolsets.

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Generated by 🔍 GitHub MCP Structural Analysis · agent · 111.2 AIC · ⌖ 43.7 AIC · ⊞ 11.8K · ◷

• expires on Aug 13, 2026, 4:18 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 #52494.