[mcp-analysis] MCP Structural Analysis - 2026-10-09

[github-actions[bot]]

Tested 10 representative GitHub MCP tools across all available toolsets (plus the injected workflow context block). Average usefulness today: 3.1/5, consistent with the 5-day rolling average (2.6–3.2). Best: github_context (5/5, zero MCP overhead). Worst: search_code (1/5, hit a secondary rate limit for the 4th time in 5 runs). The dominant structural issue remains unchanged: single-item/small-list MCP responses carry a fixed ~2,300-byte base64 server-icon block in _meta that often outweighs the actual payload 4–6x, while the mcpscripts CLI passthrough (list_workflows) and the injected context block avoid it entirely.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	3,903
Average Usefulness Rating	3.1/5
Best Rated Tool	github_context: 5/5
Worst Rated Tool	search_code: 1/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Excellent — zero-cost identity metadata, no MCP call needed
list_pull_requests	pull_requests	⭐⭐⭐⭐⭐	Excellent — minimal, fully actionable record with fields filter
get_file_contents	repos	⭐⭐⭐⭐	Good — directory metadata listing surfaced large-file sizes cleanly
list_workflows (mcpscripts)	actions	⭐⭐⭐⭐	Good — clean pagination envelope, no icon-metadata bloat
list_issues	issues	⭐⭐⭐	Adequate — clean schema, but content withheld by integrity policy this call
list_discussions	discussions	⭐⭐⭐	Adequate — usable but ~80% of bytes are icon-metadata bloat
get_label	labels	⭐⭐	Limited — tiny real payload dwarfed ~6x by icon metadata
list_code_scanning_alerts	code_security	⭐⭐	Limited — all 30 matching alerts withheld by integrity policy, zero signal
search_users	users	⭐⭐	Limited — thin fields, needs follow-up call, icon overhead dominates
search_code	search	⭐	Poor — hit secondary rate limit (429) on a minimal query

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
github_context	markdown	1	actor, repository, workspace, workflow-run-id	Flat bullet list, no MCP envelope
get_file_contents	array	1	name, path, size, type	Flat per-entry objects; fields param trims to exactly what's needed
list_issues	array	1	number, title, state, labels	Empty this call — integrity filter withheld the 1 matching issue
list_pull_requests	array	1	number, title, state, html_url	Flat, no nested user/label objects when fields is set
list_workflows	object	2	total_count, per_page, page, workflows	Nested workflows array, but shallow and consistent
list_code_scanning_alerts	array	1	(none returned)	Schema not observable this call — all items filtered
list_discussions	object	3	discussions, pageInfo, totalCount, category, user	Nested category/user objects inside each discussion
get_label	object	1	name, color, description, id	Flat, but wrapped in _meta/content MCP envelope
search_users	object	2	total_count, incomplete_results, items	items array of thin user stubs
search_code	error	1	message	Error-shaped response, no retry-after field

Response Size Analysis

Toolset	Avg Tokens (today)	Avg Tokens (5-day)	Tools Tested
repos	1,037	5,807.2	1
discussions	688	774.0	1
search	634	692.4	1
labels	614	692.4	1
users	637	848.8	1
actions	118	161.2	1
issues	44	38.8	1
pull_requests	40	63.4	1
code_security	26	6,789.8	1
workflow_context	65	80.2	1

Note: code_security's 5-day average (6,789.8) is skewed by an earlier run that returned full alert payloads before the integrity filter began withholding content — today's 26-token response reflects zero disclosed alerts, not a genuinely small payload.

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	65	markdown	⭐⭐⭐⭐⭐	Injected block, no MCP round-trip
get_file_contents	repos	1,037	array	⭐⭐⭐⭐	fields filter kept it flat; correctly flagged a 34MB WASM file and 12MB linters file by size
list_issues	issues	44	array	⭐⭐⭐	1 matching issue withheld by integrity policy — schema fine, content inaccessible
list_pull_requests	pull_requests	40	array	⭐⭐⭐⭐⭐	Clean, minimal, actionable
list_workflows	actions	118	object	⭐⭐⭐⭐	mcpscripts CLI wrapper, no icon bloat
list_code_scanning_alerts	code_security	26	array	⭐⭐	30 alerts withheld by integrity policy
list_discussions	discussions	688	object	⭐⭐⭐	Icon-metadata bloat ~80% of bytes
get_label	labels	614	object	⭐⭐	Icon-metadata bloat ~6x real payload
search_users	users	637	object	⭐⭐	Thin fields, icon bloat
search_code	search	634	error	⭐	Secondary rate limit (429)

30-Day Trend Summary

Metric	Value
Data Points	50 (5 days × 10 tools)
Average Daily Tokens	15,948.2
Average Rating Trend	Stable (2.9 → 3.2 → 2.6 → 3.0 → 3.1)

All-time average rating by toolset: workflow_context 5.0, pull_requests 4.6, actions 4.0, repos 3.0, discussions 3.0, issues 2.4, code_security 2.0, labels 2.0, users 2.0, search 1.6.

Recommendations

• High-value tools (rating 4–5): github_context (free), list_pull_requests, list_workflows via the mcpscripts CLI wrapper, get_file_contents with a fields filter for directory metadata.
• Tools needing improvement: search_code (persistent rate-limiting across 4 of 5 runs — avoid as a first-attempt exploratory call; consider search_repositories/list_commits instead where possible), get_label and search_users (icon-metadata overhead dwarfs payload on every single-item/thin-list call observed so far).
• Context-efficient tools (low tokens, high rating): github_context, list_pull_requests, list_workflows (mcpscripts).
• Context-heavy tools: get_file_contents on full directory listings (1,037 tokens even with fields filtering — scope with path_filter or target specific subdirectories when the repo tree is large), and any single-item MCP tool call carrying the recurring ~2,300-byte icon-metadata block.
• Policy note: Two calls this run (list_issues, list_code_scanning_alerts) returned structurally empty results because matching items were withheld by an integrity policy filter rather than because no matching data existed — this is expected sandbox behavior, not an MCP tool defect, but it means usefulness ratings for those two calls reflect this run's access level rather than the tool's general design.

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Generated by 🔍 GitHub MCP Structural Analysis · claude · agent · 122.4 AIC · ⊞ 8K · ◷

• expires on Oct 10, 2026, 4:04 AM UTC-08:00