[mcp-analysis] MCP Structural Analysis — 2026-05-12

[github-actions[bot]]

Quantitative + qualitative analysis of 12 GitHub MCP tools (one per toolset) measuring response size, schema shape, and usefulness for agentic work. Today's averages: 14,730 total tokens, 3.92 / 5 mean usefulness. Standouts: list_code_scanning_alerts, list_discussions, list_commits, and search_repositories all hit 5/5 — lean, well-paginated, and immediately actionable. The two heavy responses are get_file_contents (7.5K tokens for this repo's README — scales with file size) and list_workflows (2.3K tokens; per_page=1 was not honored, returning 30 workflows). get_me failed in this environment (403), reducing its observed usefulness to 1.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	12
Total Tokens (Today)	14,730
Average Usefulness Rating	3.92 / 5
Best Rated Tools	list_code_scanning_alerts, list_discussions, list_commits, search_repositories (5/5)
Worst Rated Tool	get_me (1/5 — 403 in this environment)
Largest Response	get_file_contents (~7,500 tokens for README)
Smallest Useful Response	list_code_scanning_alerts (~10 tokens, empty array)

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
list_code_scanning_alerts	code_security	⭐⭐⭐⭐⭐	Tightest signal-to-noise; empty array when nothing to report
list_discussions	discussions	⭐⭐⭐⭐⭐	Includes pageInfo cursors + totalCount; lean per-item payload
list_commits	commits	⭐⭐⭐⭐⭐	Compact, author/committer embedded, pagination respected
search_repositories	search	⭐⭐⭐⭐⭐	minimal_output=true default — ideal for agent triage
get_file_contents	repos	⭐⭐⭐⭐	Complete content + SHA + URI; size scales with file
list_issues	issues	⭐⭐⭐⭐	Clean shape with embedded user/labels; body drives size
list_pull_requests	pull_requests	⭐⭐⭐⭐	Base/head refs + embedded user; good for triage
list_releases	releases	⭐⭐⭐⭐	Complete release metadata; body markdown can be large
list_branches	branches	⭐⭐⭐⭐	Minimal but lacks last-commit metadata — supplementary call often needed
list_workflows	actions	⭐⭐⭐	Useful registry but per_page=1 not honored; redundant URL variants bloat size
list_label	labels	⭐⭐⭐	Returns all labels with no pagination — heavy on label-rich repos
get_me	context	⭐	Failed with 403 in workflow env; not useful when integration lacks user scope

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	error envelope	2	isError, content	Error path returns wrapped content array
get_file_contents	object	3	content, resource.mimeType, resource.text, resource.uri	Nests file inside resource envelope
list_issues	array	3	number, title, state, labels, user, body	Embeds user + labels inline — no extra calls needed
list_pull_requests	array	4	number, title, state, base, head, labels, user, merged	Deepest of the listing tools due to head/base nesting
list_workflows	object	3	total_count, workflows[].id/name/path/state/html_url	Wraps array in totals envelope
list_code_scanning_alerts	array	1	— (empty)	Schema-clean even when empty
list_discussions	object	3	discussions[], pageInfo, totalCount	GraphQL-style cursors stand out as a positive pattern
list_label	object	2	labels[].color/description/id/name, totalCount	Flat per item; no pagination knobs
list_commits	array	3	sha, commit.message, author, committer, html_url	Author + committer separate but flat
list_releases	array	3	tag_name, name, prerelease, body, published_at, author	Body drives variance
list_branches	array	1	name, sha, protected	Smallest schema observed
search_repositories	array	1	full_name, description, language, stargazers_count, topics, forks_count	Flat, ideal for triage

Response Size Analysis

Toolset	Avg Tokens	Tools Tested
repos	7,500	get_file_contents
actions	2,300	list_workflows
labels	1,750	list_label
issues	1,400	list_issues
pull_requests	880	list_pull_requests
releases	380	list_releases
commits	180	list_commits
search	130	search_repositories
discussions	130	list_discussions
branches	40	list_branches
context	30	get_me
code_security	10	list_code_scanning_alerts

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	30	error	⭐	403 in this environment — fails without user scope
get_file_contents	repos	7,500	object	⭐⭐⭐⭐	Heavy for large files; consider asking for path filters when available
list_issues	issues	1,400	array	⭐⭐⭐⭐	Body length dominates payload size
list_pull_requests	pull_requests	880	array	⭐⭐⭐⭐	base/head adds depth but adds real signal
list_workflows	actions	2,300	object	⭐⭐⭐	per_page ignored — returns ~30; URL fields are redundant
list_code_scanning_alerts	code_security	10	array	⭐⭐⭐⭐⭐	Empty array is gold for negative-signal queries
list_discussions	discussions	130	object	⭐⭐⭐⭐⭐	GraphQL cursors enable efficient pagination
list_label	labels	1,750	object	⭐⭐⭐	No perPage honored — returns all 100+ labels
list_commits	commits	180	array	⭐⭐⭐⭐⭐	Minimal, complete, paginated cleanly
list_releases	releases	380	array	⭐⭐⭐⭐	Watch for long release notes
list_branches	branches	40	array	⭐⭐⭐⭐	Tiny; needs supplementary call for last commit
search_repositories	search	130	array	⭐⭐⭐⭐⭐	minimal_output=true default is exemplary

30-Day Trend Summary

Metric	Value
Data Points	12 (first day — baseline)
Average Daily Tokens	14,730
Average Rating Trend	Baseline (no prior data)

Future runs will append to /tmp/gh-aw/cache-memory/mcp_analysis.jsonl for rolling 30-day trend tracking.

Recommendations

• High-value tools (rating 4–5): list_code_scanning_alerts, list_discussions, list_commits, search_repositories, get_file_contents, list_issues, list_pull_requests, list_releases, list_branches
• Tools needing improvement: list_workflows (honor per_page; trim redundant URL fields), list_label (add pagination — currently returns the entire label set), get_me (verify required scopes / surface clearer error envelope)
• Context-efficient tools (low tokens, high rating): list_code_scanning_alerts, list_branches, list_discussions, search_repositories, list_commits
• Context-heavy tools: get_file_contents (size = file size), list_workflows (pagination ignored), list_label (no pagination)
• Schema patterns to favor: GraphQL-style pageInfo + totalCount (as in list_discussions) and minimal_output toggles (as in search_repositories)

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References:

• §25734458900

Generated by GitHub MCP Structural Analysis · ● 14.4M · ◷

• expires on May 13, 2026, 12:44 PM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #31939.