[mcp-analysis] MCP Structural Analysis - 2026-06-03

[github-actions[bot]]

Structural + quantitative analysis of GitHub MCP tool responses, measuring response size (tokens), schema shape, and usefulness for autonomous agentic work. 8 toolsets sampled today with minimal-parameter calls.

Key Findings

• 8 tools analyzed across 8 toolsets — total 5,684 tokens, average usefulness 3.62 / 5.
• Best for agents (⭐⭐⭐⭐⭐): get_file_contents and search_repositories — complete, flat, no envelope bloat.
• Most context-heavy: list_issues (~2,400 tok) and list_code_scanning_alerts (~1,450 tok) — both inflated by content, not schema. Code-scanning embeds the full rule help text + code examples in every alert.
• Lowest value today: get_me returned 403 Resource not accessible by integration — unavailable to this run's token (normally a 5/5 tool).
• Pattern confirmed: token cost is overwhelmingly content-driven (large issue/PR bodies, embedded help text), not schema overhead. Compact list schemas (list_discussions, search_repositories) stay lean.

Recommendations

• Prefer for discovery: search_repositories (180 tok, all signals) and list_discussions (280 tok, cursor-paginated) — best context-per-value.
• Use list_* over search_* for issues: list_issues ships compact label names; search_issues carries reactions + ~10 *_url fields of envelope bloat (per prior runs).
• Budget for list_code_scanning_alerts: request narrow severity and small pages — each alert carries a verbose embedded rule help body that dominates the response.
• Avoid get_me reliance in restricted-integration contexts; derive identity from the workflow's GitHub context instead.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	8
Total Tokens (Today)	5,684
Average Usefulness Rating	3.62 / 5
Best Rated Tools	get_file_contents, search_repositories: 5/5
Worst Rated Tool	get_me: 1/5 (403 — unavailable to this integration)

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
get_file_contents	repos	⭐⭐⭐⭐⭐	Full text + SHA + URI in one call; immediately actionable
search_repositories	search	⭐⭐⭐⭐⭐	Flat minimal_output, all discovery signals, zero URL bloat
list_issues	issues	⭐⭐⭐⭐	Compact labels/user; cost driven by large issue body
list_pull_requests	pull_requests	⭐⭐⭐⭐	Rich PR-ops data (base/head ref+sha, draft/merged)
list_discussions	discussions	⭐⭐⭐⭐	Lean items + cursor pagination + totalCount
list_code_scanning_alerts	code_security	⭐⭐⭐	Actionable location/severity but verbose embedded help text
get_label	labels	⭐⭐⭐	Complete but low standalone value (needs exact name)
get_me	context	⭐	403 Resource not accessible by integration this run

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	object	1	error	403 error envelope only
get_file_contents	object	2	content, sha, uri, mimeType	Clean resource wrapper
list_issues	array	3	number, title, state, labels, user, timestamps, body	Labels as plain names
list_pull_requests	array	4	number, title, draft, merged, base, head, user	Depth from base/head.repo
list_code_scanning_alerts	array	4	number, rule, tool, state, most_recent_instance, location	Full help text per alert
list_discussions	object	3	discussions, category, user, pageInfo, totalCount	Cursor pagination
get_label	object	1	name, color, description, id	Flat, tiny
search_repositories	array	2	full_name, language, stargazers_count, topics, open_issues_count	Flat discovery signals

Response Size Analysis

Toolset	Avg Tokens	Tools Tested
issues	2,400	list_issues
code_security	1,450	list_code_scanning_alerts
repos	660	get_file_contents
pull_requests	620	list_pull_requests
discussions	280	list_discussions
search	180	search_repositories
labels	70	get_label
context	24	get_me (error)

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	24	object	1	403 not accessible by integration; normally 5/5
get_file_contents	repos	660	object	5	File text + SHA + URI; cost scales with file size
list_issues	issues	2,400	array	4	Compact schema; large auto-generated body inflated cost
list_pull_requests	pull_requests	620	array	4	base/head ref+sha, draft/merged flags — branch-op ready
list_code_scanning_alerts	code_security	1,450	array	3	Embeds full rule help + Go examples per alert (bloat)
list_discussions	discussions	280	object	4	Lean list + cursor pagination, totalCount=6868
get_label	labels	70	object	3	Minimal complete object; low standalone value
search_repositories	search	180	array	5	Best-in-class context efficiency

30-Day Trend Summary

Metric	Value
Data Points	99
Days Covered	10
Avg Tokens / Tool (30d)	1,321.8
Rating Trend	Stable

Coverage Notes

• actions toolset (list_workflows) was not sampled — the tool's permission was not granted to this run; recorded as a coverage gap rather than a measured data point.
• get_me (context) returned a 403; counted as a measured-but-unavailable result.

Recommendations (detail)

• High-value tools (4–5): get_file_contents, search_repositories, list_issues, list_pull_requests, list_discussions.
• Context-efficient (low tokens, high rating): search_repositories (180), list_discussions (280).
• Context-heavy (high tokens): list_issues (2,400), list_code_scanning_alerts (1,450).
• Needs caution: list_code_scanning_alerts (per-alert help-text bloat); get_me (integration-restricted); get_label (low standalone value vs list_label).

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Average Tokens per Tool

Size vs Usefulness

References: §26887564237

Generated by 🔍 GitHub MCP Structural Analysis · opus48 8.4M · ◷

• expires on Jun 4, 2026, 1:31 PM UTC

[github-actions[bot]]

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

A newer discussion is available at Discussion #36907.