[mcp-analysis] MCP Structural Analysis - 2026-10-21

[github-actions[bot]]

Tested 11 representative tool calls across 10 GitHub MCP toolsets plus the injected workflow-context block. Ratings remained stable versus the 7-day history (daily average 3.0-3.4/5). The cheapest, most reliable sources continue to be the workflow_context block, the mcpscripts wrapper tools (list_workflows, list_label), and explicit-fields calls like list_pull_requests. The search_code tool hit a 429 rate limit again, and four toolsets (issues, discussions, code_security, and one search sample) returned responses with their sole result removed by integrity/secrecy policy filtering rather than real data.

Key Metrics (Today)

Metric	Value
Tools Analyzed	11
Total Tokens (Today)	3,606
Average Usefulness Rating	3.1/5
Best Rated Tools	github_context, list_pull_requests, list_label, list_workflows (5/5 each)
Worst Rated Tool	search_code: 1/5 (429 rate limited)

Full Structural Analysis Report

Usefulness Ratings for Agentic Work (Today)

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Excellent, zero MCP overhead
list_pull_requests	pull_requests	⭐⭐⭐⭐⭐	Excellent, explicit fields keep payload clean
list_label (wrapper)	labels	⭐⭐⭐⭐⭐	Excellent, flat JSON, no icon/meta bloat
list_workflows (wrapper)	actions	⭐⭐⭐⭐⭐	Excellent, flat JSON, explicit pagination
get_file_contents	repos	⭐⭐⭐	Adequate, requires size pre-check, body duplicated
list_issues	issues	⭐⭐	Limited, sole result integrity-filtered
list_code_scanning_alerts	code_security	⭐⭐	Limited, sole result secrecy-filtered
list_discussions	discussions	⭐⭐	Limited, sole result integrity-filtered
search_users	users	⭐⭐	Limited, useful payload buried in icon/meta overhead
search_issues	search	⭐⭐	Limited, sole result integrity-filtered
search_code	search	⭐	Poor, 429 rate limited

Schema Analysis (Today)

Tool	Type	Depth	Key Fields	Notes
github_context	markdown	1	actor, repository, workspace, workflow-run-id	Injected, no MCP call
get_file_contents	object	3	content[].text, resource.text, structuredContent	Body duplicated in two places
list_issues	array	1	(filtered)	Integrity policy removed sole item
list_pull_requests	array	1	number, title, state, html_url	Clean flat array
list_code_scanning_alerts	array	1	(filtered)	Secrecy policy removed sole item
list_discussions	array	1	(filtered)	Integrity policy removed sole item
list_label (wrapper)	object	2	labels[].name/color/description, item_count	No icon/meta wrapper
search_users	object	2	items[].login/id/avatar_url, total_count	Fixed ~2.4KB icon/meta overhead
search_code	object (error)	1	error: 429	Rate limited
search_issues	array	1	(filtered)	Integrity policy removed sole item
list_workflows (wrapper)	object	2	workflows[].name/path/state, total_count	No icon/meta wrapper

Response Size Analysis (30-Day Toolset Averages)

Toolset	Avg Tokens	Tools Tested
repos	5,035	8
users	828	8
labels	621	8
discussions	532	8
search	456	12
actions	162	8
issues	154	8
pull_requests	143	8
code_security	76	8
workflow_context	66	8

Tool-by-Tool Analysis (Today)

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	58	markdown	⭐⭐⭐⭐⭐	Zero overhead identity block
get_file_contents	repos	1,380	object (depth 3)	⭐⭐⭐	Pre-checked via directory listing before reading LICENSE
list_issues	issues	65	array	⭐⭐	Sole issue (#70858) integrity-filtered
list_pull_requests	pull_requests	40	array	⭐⭐⭐⭐⭐	Open PR #70865, fully actionable
list_code_scanning_alerts	code_security	70	array	⭐⭐	Sole alert secrecy-filtered
list_discussions	discussions	68	array	⭐⭐	Sole discussion integrity-filtered
list_label (wrapper)	labels	106	object	⭐⭐⭐⭐⭐	Clean 425-char response; sampled label name itself was the unusual string [[optimization
search_users	users	682	object	⭐⭐	pelikhan found, but buried in icon/meta bloat
search_code	search	900	object (error)	⭐	429, retry ~33s
search_issues	search	65	array	⭐⭐	Sole match (#70848) integrity-filtered
list_workflows (wrapper)	actions	172	object	⭐⭐⭐⭐⭐	CI workflow, 687-char clean payload

30-Day Trend Summary

Metric	Value
Data Points	84
Distinct Sampled Days	8 (2026-10-12 through 2026-10-21, weekends skipped)
Average Daily Tokens	8,300
Average Rating Trend	Stable (daily average ranges 3.0-3.4/5, today 3.1/5)

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Recommendations

• High-value tools (30-day avg rating 4-5): list_workflows/list_label via the mcpscripts wrapper, github_context (zero-cost), and list_pull_requests with explicit fields. Prefer these patterns whenever available.
• Tools needing improvement: issues, discussions, code_security, and one of the two search samples each hit the same failure mode on nearly every sampled day: the single queried item is removed by integrity or secrecy policy filtering before reaching the agent, yielding an empty array instead of real data. This is now a stable systemic gap, not noise, and agents relying on single-item exploratory queries against these toolsets should expect empty results and plan a fallback (broader queries or a different item).
• search_code has hit a secondary GitHub rate limit (429) on the majority of sampled days. Treat it as unreliable and fall back to search_issues/search_pull_requests with narrower scope, or add client-side backoff.
• Context-efficient, high-value tools: github_context (66 avg tokens), the list_workflows wrapper (162 avg tokens), and list_pull_requests (143 avg tokens) all combine low token cost with 5/5 or near-5/5 ratings.
• Context-heavy tools: repos/get_file_contents (5,035 avg tokens, driven by full-file reads and one oversized 26,901-token failure on 2026-10-19) and users/search_users (828 avg tokens, mostly fixed icon/meta overhead) deliver the least value per token. The AGENTS.md large-file guard (pre-check via directory listing before reading) is confirmed as the correct mitigation for repos; no equivalent mitigation exists yet for the search_users icon/meta overhead since no dedicated get_user tool is available.

Generated by 🔍 GitHub MCP Structural Analysis · claude · agent · 92.4 AIC · ⌖ 1 AIC · ⊞ 8.1K · ◷

• expires on Oct 22, 2026, 4:04 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 #71128.