[mcp-analysis] MCP Structural Analysis - 2026-07-07

[github-actions[bot]]

GitHub MCP Structural Analysis — 2026-07-07

Analyzed 10 representative tools across 10 toolsets for response size, schema structure, and usefulness for agentic work. Data appended to a 17-day rolling window (156 points).

Key findings:

• Average usefulness: 3.8/5 (up from 3.22 at window start — steady improvement as measurement conventions stabilized).
• Best rated (5/5): get_file_contents, get_label, list_discussions, search_repositories — flat/compact schemas, every field actionable, bounded cost.
• Worst rated: get_me (1/5) — persistent 403 Resource not accessible by integration under the CI workflow token; agents must read identity from workflow context, not this tool.
• Heaviest responses: list_code_scanning_alerts (2635t) and list_issues (1860t) — both bloated by inlined content (full rule.help markdown / full issue bodies), not schema breadth.
• Schema pattern: the two cheapest-yet-complete tools (get_label 40t d1, search_repositories 116t d2) are flat; cost and confusion both rise with nesting (list_pull_requests d4 re-embeds a repo object under both base and head; list_code_scanning_alerts d5).

Recommendations (always visible):

• Prefer for agentic loops: list_discussions (only tool with real GraphQL cursor pagination + totalCount), search_repositories/search_users (always set perPage), get_label, actions_list.
• Use with care / post-process: list_issues and list_pull_requests inline full bodies — request a single item and strip body before reasoning, or fetch detail on-demand.
• list_code_scanning_alerts: finding + location fields are excellent, but the per-result full rule.help markdown is ~1255 chars/alert of static doc — treat help as an on-demand lookup, not per-row payload.
• get_me is unusable in CI identity — do not call it; read actor from the workflow's github-context.
• Dispatcher gotcha: actions_list requires a method arg (e.g. list_workflows); bare list_workflows is not exposed. search_users requires query (not q).

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	5,969
Average Usefulness Rating	3.8/5
Best Rated Tool	get_file_contents / get_label / list_discussions / search_repositories: 5/5
Worst Rated Tool	get_me: 1/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
get_file_contents	repos	⭐⭐⭐⭐⭐	SHA-pinned resource + full text; safe follow-up edits
get_label	labels	⭐⭐⭐⭐⭐	Flat, complete, best token-efficiency
list_discussions	discussions	⭐⭐⭐⭐⭐	Only real cursor pagination + totalCount
search_repositories	search	⭐⭐⭐⭐⭐	All discovery signals, flat, bounded by perPage
actions_list	actions	⭐⭐⭐⭐	Clean envelope; needs method arg; 4 redundant URL fields
list_pull_requests	pull_requests	⭐⭐⭐⭐	Rich flags; base+head each re-embed repo (d4); body inlined
search_users	users	⭐⭐⭐⭐	Compact envelope; needs 'query'; no deep cursor
list_issues	issues	⭐⭐⭐	Crisp triage fields but full body dumped (~6.9KB)
list_code_scanning_alerts	code_security	⭐⭐	Great findings, but full rule.help inlined per alert
get_me	context	⭐	403 under CI integration token; unusable here

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_label	object	1	name, color, description, id	Flat, ideal
search_repositories	array	2	full_name, language, stars, forks, open_issues, topics	Flat discovery signals
get_file_contents	object	3	resource.text, mimeType, (redacted) uri (SHA-pinned)	text + SHA note
list_discussions	object	3	discussions[], pageInfo{endCursor,hasNextPage}, totalCount	GraphQL cursor pagination
search_users	object	3	total_count, items[login,id,profile_url,avatar_url]	Search envelope
actions_list	object	3	total_count, workflows[id,name,path,state]	+url/html_url/badge_url/node_id bloat
list_issues	array	3	number, title, state, body, labels, user	body inlined (~6912 chars)
list_pull_requests	array	4	number, draft, merged, base{ref,sha}, head{ref,sha}, reviewers	base+head re-embed repo
get_me	error	2	content, isError	403 integration token
list_code_scanning_alerts	array	5	rule{id,security_severity_level,help}, most_recent_instance.location	full help markdown/alert

Response Size Analysis

Toolset	Avg Tokens	Tools Tested
code_security	2635	list_code_scanning_alerts
issues	1860	list_issues
pull_requests	547	list_pull_requests
repos	344	get_file_contents
discussions	150	list_discussions
actions	131	actions_list
search	116	search_repositories
users	108	search_users
labels	40	get_label
context	38	get_me

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	38	error d2	1	403 Resource not accessible by integration; read actor from context
get_file_contents	repos	344	object d3	5	LICENSE as SHA-pinned resource + full text; scales with file size
list_issues	issues	1860	array d3	3	Full ~6.9KB AI-audit body inlined for one issue; triage fields crisp
list_pull_requests	pull_requests	547	array d4	4	draft/merged/base/head/reviewers; base+head re-embed repo object
actions_list	actions	131	object d3	4	Needs method=list_workflows; total_count:331; redundant URL fields
list_code_scanning_alerts	code_security	2635	array d5	2	6 open alerts; each inlines ~1255-char rule.help markdown
list_discussions	discussions	150	object d3	5	Cursor pagination + totalCount=8161 (+47/day); no body bloat
get_label	labels	40	object d1	5	Flat name/color/description/id + node id
search_users	users	108	object d3	4	Needs 'query'; total_count=2 + minimal item shape
search_repositories	search	116	array d2	5	All signals; stars 4724->4727, open_issues 212->240

30-Day Trend Summary

Metric	Value
Data Points	156
Days Covered	17 (2026-06-08 → 2026-07-07)
Average Daily Tokens	7,515
Average Rating Trend	Improving (3.22 → 3.80)

Trend notes: discussions totalCount 8114→8161 (~+47/day); repo stars 4724→4727; open issues 212→240 (+28); code-scanning payload eased 2839→2635 as one alert cleared. The two heavy tools (issues, code_security) remain the dominant cost drivers every run — cost is content-inlining, not schema breadth.

Recommendations

• High-value tools (rating 4-5): get_file_contents, get_label, list_discussions, search_repositories, actions_list, list_pull_requests, search_users.
• Tools needing improvement: list_code_scanning_alerts (move rule.help to on-demand), list_issues (excerpt bodies in list mode), get_me (403 in CI — n/a).
• Context-efficient (low tokens, high rating): get_label (40t), search_users (108t), search_repositories (116t), actions_list (131t), list_discussions (150t).
• Context-heavy (high tokens): list_code_scanning_alerts (2635t), list_issues (1860t), list_pull_requests (547t).

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Warning

Firewall blocked 1 domain

The following domain was blocked by the firewall during workflow execution:

• awmgmcpg

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "awmgmcpg"

See Network Configuration for more information.

Generated by 🔍 GitHub MCP Structural Analysis · 240.3 AIC · ⌖ 16.5 AIC · ⊞ 9.6K · ◷

• expires on Jul 8, 2026, 4:58 AM UTC-08:00

[github-actions[bot]]

Caveman smoke wave. Tests run. Rock on.

Warning

Firewall blocked 6 domains

The following domains were blocked by the firewall during workflow execution:

• accounts.google.com
• android.clients.google.com
• clients2.google.com
• contentautofill.googleapis.com
• safebrowsingohttpgateway.googleapis.com
• www.google.com

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "accounts.google.com"
    - "android.clients.google.com"
    - "clients2.google.com"
    - "contentautofill.googleapis.com"
    - "safebrowsingohttpgateway.googleapis.com"
    - "www.google.com"

See Network Configuration for more information.

📰 BREAKING: Report filed by Smoke Copilot · 73.8 AIC · ⌖ 3.43 AIC · ⊞ 19K · ◷
_{Comment /smoke-copilot to run again
Add label smoke to run again}