[mcp-analysis] MCP Structural Analysis - 2026-10-20

[github-actions[bot]]

Overview

Tested 10 representative GitHub MCP tools/toolsets with minimal parameters (perPage: 1 or single-item queries) and recorded size, schema, and agentic usefulness. Today's run used 3,930 tokens across the sample, with an average usefulness rating of 3.40/5 — the highest of the 7 sampled days in the 30-day window (2026-10-12 to 2026-10-20, 73 data points).

The standout finding: search_issues returned a real, unfiltered closed issue with a full markdown body for the first time in 3 sampled days (prior days hit integrity-policy filtering), lifting its rating from 1-2 to 4/5. Meanwhile list_issues, list_code_scanning_alerts, and list_discussions continue a now 6-day streak of single-item empty-array responses due to integrity/secrecy policy filtering — a structural limitation, not a transient blip.

Key Metrics

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	3,930
Average Usefulness Rating	3.40/5
Best Rated Tools	github_context, list_pull_requests, list_workflows (5/5 each)
Worst Rated Tools	list_issues, list_code_scanning_alerts, list_discussions, search_users (2/5 each)

Full Structural Analysis Report

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Excellent — zero MCP overhead, directly actionable identity fields
list_pull_requests	pull_requests	⭐⭐⭐⭐⭐	Excellent — fields-filtered, flat, zero bloat
list_workflows (mcpscripts)	actions	⭐⭐⭐⭐⭐	Excellent — clean JSON, explicit pagination, zero icon/_meta bloat
search_issues	search	⭐⭐⭐⭐	Good — full real content this run, but no body-truncation option
get_label	labels	⭐⭐⭐⭐	Good payload, but fixed icon overhead dwarfs it
get_file_contents	repos	⭐⭐⭐	Adequate — requires pre-check of file size before reading
list_issues	issues	⭐⭐	Limited — result policy-filtered, false-negative risk
list_code_scanning_alerts	code_security	⭐⭐	Limited — result policy-filtered, false-negative risk
list_discussions	discussions	⭐⭐	Limited — result policy-filtered, false-negative risk
search_users	users	⭐⭐	Limited — no dedicated get_user tool, payload buried in overhead

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
github_context	markdown	1	actor, repository, workspace, workflow-run-id	Prompt-injected, no MCP call
get_file_contents	object	3	content[].text, resource.text, resource.uri	Body duplicated twice in response
list_issues	array	1	(empty — filtered)	Integrity-policy gated
list_pull_requests	array	1	number, title, state, html_url	Clean flat record
list_code_scanning_alerts	array	1	(empty — filtered)	Secrecy-policy gated
list_discussions	array	1	(empty — filtered)	Integrity-policy gated
get_label	object	1	color, description, name	Flat, but icon overhead dominates
search_users	object	2	items[].login, items[].id, total_count	No dedicated get_user substitute
search_issues	object	2	number, title, body, labels, reactions	Full body included, no truncation
list_workflows (mcpscripts)	object	2	total_count, workflows[].name/path/state	Explicit pagination metadata

Response Size Analysis (30-day average by toolset)

Toolset	Avg Tokens	Tools Tested
repos	5,557	7
users	849	7
labels	694	7
discussions	598	7
search	451	7
issues	167	7
actions	160	7
pull_requests	157	7
code_security	77	7
workflow_context	67	7

Note: repos average is inflated by a one-off 26,901-token oversized-README failure on 2026-10-19; excluding that outlier the toolset averages ~1,400 tokens (consistent with the LICENSE-file baseline).

Tool-by-Tool Analysis (Today)

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	58	markdown	⭐⭐⭐⭐⭐	Cheapest high-value source, 7th straight day
get_file_contents	repos	1,380	object	⭐⭐⭐	Directory-listing precheck (988 tok) avoided an oversized README read
list_issues	issues	64	array	⭐⭐	Policy-filtered empty array, 6th straight day
list_pull_requests	pull_requests	40	array	⭐⭐⭐⭐⭐	Smallest clean payload this run
list_code_scanning_alerts	code_security	70	array	⭐⭐	Policy-filtered empty array, 6th straight day
list_discussions	discussions	68	array	⭐⭐	Policy-filtered empty array
get_label	labels	616	object	⭐⭐⭐⭐	mcpscripts wrapper still strictly better (85 tok historically)
search_users	users	682	object	⭐⭐	Identical byte count to 2026-10-19 — overhead is fixed, not content-dependent
search_issues	search	780	object	⭐⭐⭐⭐	First unfiltered real result in 3 sampled days
list_workflows (mcpscripts)	actions	172	object	⭐⭐⭐⭐⭐	Exact byte count confirmed via bridge log (687 chars)

30-Day Trend Summary

Metric	Value
Data Points	73
Average Daily Tokens (excl. error outlier)	5,127
Average Rating Trend	Stable, slightly improving (3.27 → 3.30 → 3.00 → 3.18 → 3.20 → 3.00 → 3.40)

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Recommendations

• High-value tools (rating 4-5): github_context (free), list_pull_requests/list_workflows with explicit fields/mcpscripts wrappers, get_label, and search_issues when results clear integrity filtering.
• Tools needing improvement: list_issues, list_code_scanning_alerts, list_discussions are now structurally gated by integrity/secrecy policy on single-item exploratory queries in this repo — 6 consecutive sampled days of empty-array false negatives. Agents relying on these for triage should treat an empty array as "possibly filtered," not "confirmed empty." search_users has no low-overhead substitute since no get_user tool exists.
• Context-efficient tools (low tokens, high rating): github_context (58 tok), list_pull_requests (40 tok), list_workflows via mcpscripts (172 tok) — all consistently ⭐⭐⭐⭐⭐ across the 7-day sample.
• Context-heavy tools: get_file_contents/repos (1,380 tok baseline, worse if the large-file guard isn't applied first), search_users/get_label built-in calls (both >600 tok due to fixed ~2.4KB icon/_meta wrapper overhead that the mcpscripts wrappers avoid entirely).

Generated by 🔍 GitHub MCP Structural Analysis · claude · agent · 89.6 AIC · ⌖ 0.944 AIC · ⊞ 8.1K · ◷

• expires on Oct 21, 2026, 4:05 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 #70877.