[mcp-analysis] MCP Structural Analysis - 2026-09-14

[github-actions[bot]]

Analyzed 10 representative GitHub MCP tools (one per toolset) for token cost and schema quality. Today's calls totaled 22,902 tokens at an average usefulness of 3.2/5. Best: github_context (workflow-injected, 5/5, 80 tokens). Worst: search_code (1/5) — hit a 429 rate limit with no retry. Recurring pattern across 14 measured days: a fixed _meta.serverInfo icon block inflates every non-wrapper tool's payload regardless of actual content size, while the mcpscripts wrapper (list_workflows) stays clean.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools analyzed today	10
Total tokens today	22,902
Average usefulness rating	3.2 / 5
Best tool	github_context (5/5, 80 tokens)
Worst tool	search_code (1/5, rate-limited)
30-day data points	140
Avg daily total tokens (30d)	8,682.9

Usefulness Ratings

Tool	Toolset	Rating
github_context	workflow_context	⭐⭐⭐⭐⭐
list_workflows	actions	⭐⭐⭐⭐⭐
list_code_scanning_alerts	code_security	⭐⭐⭐⭐
list_discussions	discussions	⭐⭐⭐
list_pull_requests	pull_requests	⭐⭐⭐
list_issues	issues	⭐⭐⭐
search_users	users	⭐⭐⭐
get_label	labels	⭐⭐⭐
get_file_contents	repos	⭐⭐
search_code	search	⭐

Schema Analysis

Tool	Schema Type	Nesting Depth	Key Fields
github_context	markdown	1	actor, repository, workspace, workflow-run-id
list_workflows	object	2	total_count, per_page, workflows[].id/name/path/state
list_code_scanning_alerts	array	3	rule.id, rule.severity, most_recent_instance.location, state
list_discussions	object	3	discussions[].category.name, pageInfo.hasNextPage, totalCount
list_pull_requests	array	2	number, title, state, body, user.login, head.ref, base.ref
list_issues	array	2	number, title, state, labels, body, user.login
search_users	object	2	items[].login, items[].id, total_count
get_label	object	1	name, color, description, id
get_file_contents	object	3	content.resource.text, _meta.serverInfo
search_code	object	1	error message (429 rate limit)

Response Size Analysis (All-Time Avg by Toolset)

Toolset	Avg Tokens
repos	2,797
code_security	2,644
labels	882
discussions	692
users	616
search	588
actions	171
issues	100
pull_requests	100
workflow_context	92

Tool-by-Tool Analysis

Tool	Toolset	Tokens (today)	Notes
github_context	workflow_context	80	Zero-cost injected Markdown identity block; no MCP call needed.
list_workflows	actions	172	mcpscripts wrapper returns clean JSON with explicit pagination metadata and zero icon bloat — consistent for 14 days.
list_code_scanning_alerts	code_security	7,200	13 open critical/high alerts; CodeQL entries highly actionable, third-party scanner entries (Semgrep, Malicious Code Scanner) have empty rule fields and cram detail into message.
list_discussions	discussions	750	Good pagination metadata, but fixed icon _meta block dominates a perPage=1 payload.
list_pull_requests	pull_requests	900	Called without field selection; full PR body (audit summary + file table) vs. 32-48 tokens when field-filtered — fields param is the single biggest size lever.
list_issues	issues	600	Called without field selection; templated auto-generated body dominates payload.
search_users	users	630	Minimal identity fields only (login/id/avatar); adequate for disambiguation, needs a follow-up call for profile detail.
get_label	labels	610	~25 tokens of actual content wrapped in a repeated _meta icon block — per-call overhead dominates.
get_file_contents	repos	11,900	README.md grew to 45KB; auto-generated contributor list (280+ usernames) now dominates payload — largest single-call payload in 30 days.
search_code	search	60	429 rate-limited, no automatic retry — 2 of last 4 measurements have failed outright.

30-Day Trend Summary

Date	Avg Usefulness
2026-08-25	3.6
2026-08-26	3.7
2026-08-27	3.6
2026-08-28	3.6
2026-08-31	3.6
2026-09-02	3.3
2026-09-03	3.6
2026-09-04	3.6
2026-09-07	3.6
2026-09-08	3.4
2026-09-09	3.3
2026-09-10	3.6
2026-09-11	3.6
2026-09-14	3.2

Rating dipped today mainly because list_issues/list_pull_requests were called without field filters (recovering that filter would push both back to their usual 4-5 range) and search_code failed on rate limit.

Visualizations

Recommendations

High-value tools (4-5 rating): github_context (free identity source, always prefer over get_me), list_workflows via the mcpscripts wrapper (clean schema, explicit pagination — use --perPage over the built-in tool), list_code_scanning_alerts (actionable CodeQL detail, worth the token cost given state:open,severity:critical,high filtering).

Tools needing improvement: search_code needs client-side backoff/retry — it has the highest call-to-call reliability variance of any tool tested (2 of last 4 runs failed on 429). get_file_contents cost is highly target-dependent and unbounded for auto-appended files (e.g. contributor lists) — prefer range/line-limited reads where supported.

Context-efficient tools: github_context, list_workflows, list_issues/list_pull_requests (when field-filtered — omitting body recovers ~90% of the token cost seen today).

Context-heavy tools: get_file_contents (11,900 tokens, growing), list_code_scanning_alerts (7,200 tokens, inherent to multi-alert detail), both dominated less by schema depth than by the size of the underlying content being returned.

Cross-cutting fix: every non-wrapper GitHub MCP tool repeats a fixed _meta.serverInfo icon block (base64) on every response — for small-payload tools (get_label, list_discussions, search_users) this overhead is a significant fraction of total tokens. Suppressing _meta where not needed would be the single highest-leverage schema improvement available.

References:

• §34840590642

Warning

Firewall blocked 1 domain

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

• api.anthropic.com

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

network:
  allowed:
    - defaults
    - "api.anthropic.com"

See Network Configuration for more information.

Generated by 🔍 GitHub MCP Structural Analysis · claude · agent · 253.1 AIC · ⌖ 12.1 AIC · ⊞ 11.5K · ◷

• expires on Sep 15, 2026, 4:09 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 #61129.