[mcp-analysis] MCP Structural Analysis - 2026-06-25

[github-actions[bot]]

Daily structural + size analysis of 9 representative GitHub MCP tools (one per toolset) for github/gh-aw. 9 tools analyzed, total 9,250 tokens, average usefulness 3.3/5.

Key Findings

• 🟢 actions_list improved (r3 → r4): the actions toolset is now a single actions_list verb that takes a method param, and crucially perPage=1 is now honored — it returned exactly ONE workflow (~175 tokens, total_count=324). The legacy list_workflows ignored per_page and dumped 30 rows (~1,650–3,400 tokens). This is the standout efficiency win of the window.
• 🔴 get_me still r1: 6th+ consecutive 403 "Resource not accessible by integration" — the Actions integration token has no /user scope. Agents must read identity from injected context, not this tool.
• 🟡 list_code_scanning_alerts still r2: one high alert (#635 js/insecure-temporary-file) costs ~760 tokens because the full rule.help markdown (CWE-377 writeup + code samples) is inlined per result. Help belongs in on-demand lookup.
• 🟡 list_issues r3 / search_repositories r4: today's heaviest payloads — a full inlined issue body (~2,600 tok) and an unbounded 30-result repo search (~3,950 tok). Both are size-driven, not schema-driven.
• 🟢 Best-in-class small tools: get_label (~45 tok, flat), list_discussions (~200 tok, proper cursor pagination, totalCount=7,705, +41/day).

Recommendations

• High-value (r4–5): actions_list, list_pull_requests, list_discussions, get_label, get_file_contents, search_repositories — actionable, well-structured.
• Context-efficient (low tokens, high rating): get_label (45), actions_list (175), list_discussions (200).
• Context-heavy: search_repositories (pass per_page to bound results), list_issues (add a body-excluded/truncated list mode).
• Needs improvement: list_code_scanning_alerts (move rule.help out of per-result payload); get_me (non-functional under integration auth — agents should skip identity-dependent flows).

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	9
Total Tokens (Today)	9,250
Average Usefulness Rating	3.3/5
Best Rated Tool	actions_list: 4/5 (most improved)
Worst Rated Tool	get_me: 1/5 (auth-limited)

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
actions_list	actions	⭐⭐⭐⭐	Good — pagination now honored; minor URL-quad bloat
list_pull_requests	pull_requests	⭐⭐⭐⭐	Good — rich PR-ops payload; redundant nested repo objects
list_discussions	discussions	⭐⭐⭐⭐	Good — cursor pagination, very efficient; no body/labels
get_label	labels	⭐⭐⭐⭐	Good — flat & complete; narrow scope caps leverage
get_file_contents	repos	⭐⭐⭐⭐	Good — clean resource envelope; cost is file-driven
search_repositories	search	⭐⭐⭐⭐	Good schema, but unbounded 30 results inflated size
list_issues	issues	⭐⭐⭐	Adequate — full body inlined makes per-item cost unpredictable
list_code_scanning_alerts	code_security	⭐⭐	Limited — full rule.help markdown bloats every result
get_me	context	⭐	Poor — 403 under integration auth, no identity payload

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	error	1	isError, text	403 envelope only
get_label	object	1	name, color, description, id	Flat, zero bloat
list_discussions	object	3	number, title, category.name, user.login, pageInfo, totalCount	GraphQL-style, paginated
actions_list	object	2	total_count, workflows[].id/name/path/state	perPage honored; url+html_url+badge_url+node_id redundant
list_issues	array	3	number, title, state, labels, user.login, body	Full body inlined
list_pull_requests	array	4	number, state, draft, merged, base/head.ref+sha, labels, assignees	Nested repo objects redundant
list_code_scanning_alerts	array	5	number, rule.id, security_severity_level, location.path/start_line, html_url	rule.help bloat
get_file_contents	array	3	resource.text, mimeType, uri, sha	Clean; cost = file size
search_repositories	array	2	full_name, stars, forks, language, topics, open_issues_count	Flat; unbounded result count

Response Size Analysis

Toolset	Avg Tokens (today)	Tools Tested
search	3,950	search_repositories
issues	2,600	list_issues
repos	1,130	get_file_contents
code_security	760	list_code_scanning_alerts
pull_requests	360	list_pull_requests
discussions	200	list_discussions
actions	175	actions_list
labels	45	get_label
context	30	get_me

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_me	context	30	error	1	403, no /user scope under integration token
actions_list	actions	175	object	4	Now needs method param; perPage HONORED, total_count=324
list_issues	issues	2,600	array	3	Issue #40xxx AIC audit; full body dominates cost
list_pull_requests	pull_requests	360	array	4	PR #41430 version bump; light today; nested-repo bloat
list_code_scanning_alerts	code_security	760	array	2	1 high alert #635 js/insecure-temporary-file; rule.help inlined
list_discussions	discussions	200	object	4	totalCount=7,705 (+41/day); cursor pagination
get_label	labels	45	object	4	'bug' #d73a4a; flat & proportionate
get_file_contents	repos	1,130	array	4	.gitignore (SHA feb5f08); cost is file-driven
search_repositories	search	3,950	array	4	query='gh-aw' → 30 results; stars=4,693 (+4); no per_page

30-Day Trend Summary

Metric	Value
Data Points	190 rows
Days of Data	21
Average Daily Tokens	8,513
Average Rating Trend	3.73 → 3.33 (slightly declining; driven by code_security r2 + heavier issue/search payloads)
Discussions totalCount	7,705 (steady ~+41/day)
Repo stars	4,693 (+4 vs 06-24)

Recommendations

• High-value tools (rating 4–5): actions_list, list_pull_requests, list_discussions, get_label, get_file_contents, search_repositories
• Tools needing improvement: list_code_scanning_alerts (externalize rule.help), list_issues (body-excluded list mode), get_me (auth-limited in Actions context)
• Context-efficient tools (low tokens, high rating): get_label (45), actions_list (175), list_discussions (200)
• Context-heavy tools (high tokens): search_repositories (3,950 — bound with per_page), list_issues (2,600 — full body), get_file_contents (1,130 — intrinsic to file)

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References: §28170585840

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 · 222.1 AIC · ⌖ 17.2 AIC · ⊞ 11K · ◷

• expires on Jun 26, 2026, 4:47 AM UTC-08:00

[github-actions[bot]]

Test 7: discussion interaction ✅

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 - AOAI (apikey) · 169.1 AIC · ⌖ 16.8 AIC · ⊞ 17.9K · ◷

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-06-26T12:47:49.702Z.

Closed by Workflow