[mcp-analysis] MCP Structural Analysis - 2026-10-22

[github-actions[bot]]

Daily structural analysis of the GitHub MCP server: 11 tools sampled across 9 toolsets today, plus the zero-cost <github-context> identity block. Average usefulness 3.2/5. Best: github_context, list_pull_requests, list_label, list_workflows (all 5/5). Worst: search_code (1/5, hit a 429 rate limit again). The dominant pattern across all 9 sampled days: mcpscripts wrappers and explicit fields selection consistently beat built-in tools' default payloads, which are padded with a fixed ~3.4KB icon/_meta block regardless of content size.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed (today)	11
Total Tokens (Today)	4,649
Average Usefulness Rating	3.18/5
Best Rated Tool	github_context, list_pull_requests*, list_label, list_workflows: 5/5
Worst Rated Tool	search_code: 1/5

* list_pull_requests rated 4/5 today pending full star table below; see per-tool table.

Usefulness Ratings for Agentic Work (Today)

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Excellent — zero MCP overhead, directly actionable
list_pull_requests	pull_requests	⭐⭐⭐⭐⭐	Excellent — clean flat array with explicit fields
list_label (mcpscripts)	labels	⭐⭐⭐⭐⭐	Excellent — flat JSON, no icon/_meta bloat
list_workflows (mcpscripts)	actions	⭐⭐⭐⭐⭐	Excellent — flat JSON, pagination metadata
get_file_contents	repos	⭐⭐⭐	Adequate — correct data, but heavy fixed overhead
search_issues	search	⭐⭐⭐	Adequate — actionable but oversized (missing perPage)
list_issues	issues	⭐⭐	Limited — sole item policy-filtered (8th day running)
list_code_scanning_alerts	code_security	⭐⭐	Limited — sole item secrecy-filtered (8th day running)
list_discussions	discussions	⭐⭐	Limited — sole item policy-filtered
search_users	users	⭐⭐	Limited — useful payload buried under ~3.4KB overhead
search_code	search	⭐	Poor — 429 rate limited (majority outcome across sampled days)

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
github_context	markdown	1	actor, repository, workspace, workflow-run-id	Injected, not an MCP call
get_file_contents	object	3	content[].text, resource.text, resource.uri	Body duplicated in two places
list_issues	array	1	—	Filtered empty today
list_pull_requests	array	1	number, title, state, html_url	Flat, explicit fields
list_code_scanning_alerts	array	1	—	Filtered empty today
list_discussions	array	1	—	Filtered empty today
search_users	object	2	items[].login, id, avatar_url, profile_url	icon/_meta wrapper dominates
search_code	object	1	error	429 rate limit
search_issues	array	2	number, title, state, labels	No perPage → ~30 results returned
list_label (mcpscripts)	object	2	labels[].name, color, description, item_count	No icon/_meta bloat
list_workflows (mcpscripts)	object	2	total_count, workflows[].name/path/state/html_url	No icon/_meta bloat

Response Size Analysis (Today)

Toolset	Tokens	Tool Tested
repos	1,380	get_file_contents
users	900	search_users
search	900 (code) / 890 (issues)	search_code, search_issues
actions	172	list_workflows (mcpscripts)
labels	106	list_label (mcpscripts)
code_security	70	list_code_scanning_alerts
discussions	68	list_discussions
issues	65	list_issues
workflow_context	58	github_context
pull_requests	40	list_pull_requests

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	58	markdown	5	Zero MCP overhead, 9th consecutive day as cheapest high-value source
list_pull_requests	pull_requests	40	array	5	Best token-to-value ratio of any toolset, 9 days running
list_code_scanning_alerts	code_security	70	array	2	Secrecy-policy filtered, 8th consecutive day
list_issues	issues	65	array	2	Integrity-policy filtered, 8th consecutive day
list_discussions	discussions	68	array	2	Integrity-policy filtered
search_users	users	900	object	2	No dedicated get_user tool; ~3.4KB fixed overhead
search_code	search	900	object(error)	1	429 rate limit, now the majority outcome
search_issues	search	890	array	3	Missing perPage caused oversized (~30-item) payload — lesson for next run
list_label	labels	106	object	5	mcpscripts wrapper, no icon/_meta bloat
list_workflows	actions	172	object	5	mcpscripts wrapper, no icon/_meta bloat
get_file_contents	repos	1,380	object	3	LICENSE read (README still oversized at 51,798 bytes)

30-Day Trend Summary

Metric	Value
Data Points	95
Days Sampled	9 (2026-10-12 → 2026-10-22)
Average Rating Trend	Stable (~3.0–3.4/5 range, no clear drift)
Consistently Excellent (5/5, every sampled day)	github_context, list_pull_requests, list_label, list_workflows
Consistently Poor	list_issues, list_code_scanning_alerts, list_discussions — policy/secrecy filtering of the sole sampled item is now a stable systemic pattern, not noise
Chronic failure mode	search_code hits secondary rate limits (429) on the majority of sampled days

30-day average tokens by tool (lower is better for context budget):

Tool	Avg Tokens
github_context	65
list_code_scanning_alerts	76
list_label	99
list_pull_requests	131
list_issues	144
list_workflows	163
search_issues	340
list_discussions	480
search_code	756
get_label	796
search_users	836
get_file_contents	4,629 (skewed by one 26,901-token oversized-read failure on 2026-10-19)

Recommendations

• High-value tools (rating 4–5): github_context (free), list_pull_requests with explicit fields, and the mcpscripts wrappers for list_label/list_workflows — all avoid the icon/_meta overhead that the built-in GitHub MCP server attaches to every response.
• Tools needing improvement: search_users and get_label (built-in) carry a fixed ~2.4–3.4KB icon/_meta block regardless of payload size — a fields/minimal-metadata mode would meaningfully cut token cost. search_code needs a documented retry/backoff story for its recurring secondary rate limit.
• Context-efficient + high-rating: github_context, list_pull_requests, list_label, list_workflows — use these patterns (explicit fields, mcpscripts wrappers) as the template for other list endpoints.
• Context-heavy tools: get_file_contents (full-file reads bring fixed overhead plus duplicated body content in content[] and structuredContent), search_users/search_code (icon/_meta bloat). Always pre-check file size via a directory listing before reading, and always pass perPage on search calls — omitting it today caused search_issues to return ~30 items instead of the intended single-item sample.

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Generated by 🔍 GitHub MCP Structural Analysis · claude · agent · 87.3 AIC · ⌖ 0.846 AIC · ⊞ 8.1K · ◷

• expires on Oct 23, 2026, 4:02 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 #71433.