[mcp-analysis] MCP Structural Analysis - 2026-07-01

[github-actions[bot]]

GitHub MCP Structural Analysis — 2026-07-01

⚠️ Live probe did NOT run today. The GitHub MCP server failed to attach to this run: the firewall gateway returned HTTP 403 ERR_ACCESS_DENIED for (awmgmcpg/redacted) ("Server rejected the configured Authorization header"). As a result **none of the mcp__github__*` tools were available**, so no fresh size/schema/usefulness measurements could be taken. No data point was appended to the trend file (deliberately — fabricating one would corrupt the 30-day series). The analysis below is the 30-day historical view from prior successful runs; the "latest day" reflects the last good run, 2026-06-30.

Key findings (from the historical series, last good run 2026-06-30):

• 12 tools across 10 toolsets last measured; avg usefulness 4.0/5, ~7,111 tokens total that day.
• Best: search_repositories — 5/5 (117 tok, all discovery signals, no bloat). Worst: get_me — 1/5 (403 under integration token; returns an error envelope, not a profile).
• Heaviest / least efficient: list_code_scanning_alerts (~3,336 tok, 2/5) — rule.help inlines full CWE markdown + code samples per alert; and list_issues (~2,164 tok, 3/5) — full issue body always inlined even at perPage=1.
• Best value (low tokens, high rating): list_branches (30 tok, 5/5), get_label (40 tok, 5/5), search_users (65 tok, 5/5), search_repositories (117 tok, 5/5), list_discussions (151 tok, 5/5).

Recommendations

• Prefer for agentic work (4–5/5): search_repositories, list_branches, get_label, search_users, list_discussions, actions_list, get_file_contents, list_commits, list_pull_requests.
• Use sparingly / expect follow-ups: list_code_scanning_alerts (strip rule.help; heaviest tool), list_issues (needs a body-excerpt/truncation flag).
• Avoid for identity: get_me under the workflow integration token — derive actor from github-context instead.
• Infra action: investigate the recurring GitHub-MCP gateway 403 so daily live probing resumes; today's series has a gap.

Full Structural Analysis Report

Run Status

Item	Value
Analysis date	2026-07-01
Live GitHub MCP probe	❌ Failed — gateway HTTP 403 ERR_ACCESS_DENIED (Authorization header rejected)
Tools measured this run	0 (server not attached)
Data point appended	No (avoids corrupting trend)
Source of tables/charts	Historical cache (mcp_analysis.jsonl, 162 rows)

Executive Summary (last good run — 2026-06-30)

Metric	Value
Tools Analyzed	12
Total Tokens (that day)	7,111
Average Usefulness Rating	4.0/5
Best Rated Tool	search_repositories: 5/5
Worst Rated Tool	get_me: 1/5

Usefulness Ratings for Agentic Work (2026-06-30)

Tool	Toolset	Rating	Assessment
search_repositories	search	⭐⭐⭐⭐⭐	All discovery signals, no bloat
actions_list	actions	⭐⭐⭐⭐⭐	Compact workflow discovery (note method dispatch)
get_label	labels	⭐⭐⭐⭐⭐	Tiny flat lookup; node id for chaining
list_branches	repos	⭐⭐⭐⭐⭐	Minimal list + protection flag
search_users	users	⭐⭐⭐⭐⭐	Clean minimal user shape
list_discussions	discussions	⭐⭐⭐⭐⭐	Cursor pagination + totalCount; body excluded
get_file_contents	repos	⭐⭐⭐⭐	Resource envelope + sha; scales with file, no range read
list_commits	repos	⭐⭐⭐⭐	Compact; minor author-at-two-levels redundancy
list_pull_requests	pull_requests	⭐⭐⭐⭐	Rich triage; base/head each re-embed full repo
list_issues	issues	⭐⭐⭐	Full body always inlined → heavy even at perPage=1
list_code_scanning_alerts	code_security	⭐⭐	rule.help markdown inlined per alert; ~80% static docs
get_me	context	⭐	403 under integration token; error envelope, not profile

Schema Analysis (2026-06-30)

Tool	Type	Depth	Key Fields	Notes
search_repositories	array	3	full_name, language, stargazers_count, topics	Flat discovery signals
actions_list	object	3	total_count, workflows[].id/name/path/state	badge_url/node_id rarely actionable
get_label	object	1	name, color, description, id	Flat, complete
list_branches	array	2	name, sha, protected	No author/message bloat
search_users	object	3	total_count, items[].login/id/profile_url	Minimal item shape
list_discussions	object	3	discussions[].number/title/category, pageInfo, totalCount	GraphQL-style envelope
get_file_contents	object	3	content, mimeType, uri, sha	Resource envelope
list_commits	array	4	sha, commit.message, author.login	Author at two levels
list_pull_requests	array	4	number, state, base.ref, head.ref, draft, merged	Repo re-embedded twice
list_issues	array	3	number, title, state, body, user.login	Full body inlined
list_code_scanning_alerts	array	4	number, rule, most_recent_instance.location, html_url	rule.help markdown per alert
get_me	error	2	content, isError	403 error envelope

Response Size by Toolset (30-day average)

Toolset	Avg Tokens
code_security	2,156
issues	1,792
repos	1,222
pull_requests	968
actions	432
search	372
discussions	180
labels	124
users	65
context	23

Tool-by-Tool Analysis (2026-06-30)

Tool	Toolset	Tokens	Schema	Rating
list_branches	repos	30	array	5
get_label	labels	40	object	5
get_me	context	38	error	1
search_users	users	65	object	5
search_repositories	search	117	array	5
actions_list	actions	132	object	5
list_discussions	discussions	151	object	5
list_commits	repos	190	array	4
get_file_contents	repos	345	object	4
list_pull_requests	pull_requests	503	array	4
list_issues	issues	2,164	array	3
list_code_scanning_alerts	code_security	3,336	array	2

30-Day Trend Summary

Metric	Value
Data Points (rows)	162
Distinct Days	18 (2026-06-01 → 2026-06-30)
Average Daily Tokens	~7,359
Rating Trend	Stable (~4.0/5 avg)
Today's data point	Missing (gateway 403)

Recurring Patterns Observed

• Body/help inlining is the dominant token cost. list_issues and list_code_scanning_alerts are consistently the two heaviest tools because full issue bodies and full rule.help markdown are inlined regardless of pagination — a per-call truncation/opt-in flag would cut most of their footprint.
• Redundant nested repo objects in list_pull_requests (base.repo + head.repo both re-inline full repo metadata).
• Discovery/search tools are best-in-class for agentic efficiency: flat shapes, cursor pagination, no URL/permission bloat.
• get_me is structurally unusable under the integration token — agents must derive identity from workflow context.

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References: §28518514050

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 · 172.3 AIC · ⌖ 16.5 AIC · ⊞ 9.9K · ◷

• expires on Jul 2, 2026, 4:59 AM UTC-08:00

[github-actions[bot]]

Smoke ping from run 28519049107.

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 · 492.3 AIC · ⌖ 16 AIC · ⊞ 19.9K · ◷
_{Comment /smoke-copilot to run again
Add label smoke to run again}

[github-actions[bot]]

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

A newer discussion is available at Discussion #42957.