[mcp-analysis] MCP Structural Analysis - 2026-07-30

[github-actions[bot]]

MCP Structural Analysis — 2026-07-30

Analyzed 11 representative GitHub MCP tools spanning all major toolsets. Average usefulness rating: 3.0/5. Best: github_context (prompt-injected, 5/5). Worst tie: get_teams and list_label (1/5) — both effectively unusable in this run's read-only integration-token context.

Key finding: two long-running structural problems were reconfirmed live today — (1) mcp__github__get_teams remains permission-denied under the read-only token every single measurement day since tracking began, and (2) actions_list list_workflows' per_page parameter is silently ignored, returning the full 342-workflow list (~3550 tokens) regardless of the requested page size.

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	11
Total Tokens (Today)	11,146
Average Usefulness Rating	3.00/5
Best Rated Tool	github_context: 5/5
Worst Rated Tool	get_teams, list_label: 1/5

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Excellent — zero-cost prompt-injected identity metadata
list_code_scanning_alerts	code_security	⭐⭐⭐⭐	Good — rich per-finding detail (file/line/message)
list_pull_requests	pull_requests	⭐⭐⭐⭐	Good — flat, actionable with field selection
search_users	users	⭐⭐⭐⭐	Good — minimal, exact-match identity data
list_discussions	discussions	⭐⭐⭐⭐	Good — clean nesting, includes totalCount for scale
search_code	search	⭐⭐⭐	Adequate — useful snippets but nested match indices
get_file_contents	repos	⭐⭐⭐	Adequate — content usable but costly for large files
list_workflows	actions	⭐⭐	Limited — pagination param silently ignored
list_issues	issues	⭐⭐	Limited — integrity-policy redaction hides real data
get_teams	context	⭐	Poor — permission denied every run to date
list_label (untargeted)	labels	⭐	Poor — no single-label discovery path without a known name

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
github_context	markdown	1	actor, repository, workspace, workflow-run-id	Prompt block, not an MCP call
list_code_scanning_alerts	array	2	number, rule.id, rule.severity, most_recent_instance.location.path	Severity filter not strictly enforced server-side
list_pull_requests	array	1	number, title, state, html_url	Flat, field-selected
search_users	object	2	total_count, items[].login, items[].id	Flat items array
list_discussions	object	3	discussions[].category.name, pageInfo, totalCount	Cursor-based pagination present
search_code	object	3	total_count, items[].path, items[].text_matches[].fragment	Nested highlighted match fragments
get_file_contents	object	3	content[].resource.text, content[].resource.uri	Wrapped one level in resource envelope
list_workflows	object	2	total_count, workflows[].id, workflows[].path	Pagination param ignored
list_issues	array	1	(empty — redacted)	Integrity-policy filtering hides the one open issue
get_teams	error	0	—	403 permission denied
list_label	error	0	—	Returned full 820-label set, not the untargeted single label expected

Response Size Analysis (30-day average by toolset)

Toolset	Avg Tokens
repos	4,276
actions	2,035
code_security	1,767
pull_requests	1,557
issues	1,284
search	787
labels	214
discussions	183
users	134
workflow_context	76
context	37

Tool-by-Tool Analysis (Today)

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	78	markdown	5	Zero-overhead identity metadata
list_code_scanning_alerts	code_security	1,450	array	4	Rich actionable findings; severity filter loose
list_pull_requests	pull_requests	55	array	4	Field selection minimized payload well
search_users	users	45	object	4	Minimal exact-match lookup
list_discussions	discussions	145	object	4	Clean nested structure, scale context via totalCount
search_code	search	180	object	3	Useful but requires unpacking match indices
get_file_contents	repos	2,650	object	3	Cost driven by file content size, not tool overhead
list_workflows	actions	3,550	object	2	per_page param not respected — returned all 342
list_issues	issues	60	array	2	Integrity policy redacts the one real open issue
get_teams	context	33	error	1	Permission denied — consistent every run
list_label	labels	2,900	error	1	Untargeted call returns full 820-label catalog

30-Day Trend Summary

Metric	Value
Data Points	184
Average Daily Tokens	~13,607
Average Rating Trend	Stable — the same structural gaps (get_teams 403, list_workflows pagination, issue redaction) recur across all 16 measurement days in the window

Recommendations

• High-value tools (rating 4–5): github_context (free), search_users, list_pull_requests, list_discussions, list_code_scanning_alerts — prefer these when field-selection is supported.
• Tools needing improvement:
  • get_teams — permission consistently denied under the read-only integration token; agents should rely on the injected <github-context> block instead of ever attempting this call.
  • actions_list list_workflows — per_page/page params appear unenforced; expect the full workflow list regardless of pagination request, and budget context accordingly for repos with hundreds of workflows.
  • list_issues/search_issues — integrity-policy redaction silently drops matching issues from the visible array with no marker distinguishing "zero results" from "results filtered," which can mislead agents into believing no issues are open.
• Context-efficient tools (low tokens, high rating): github_context, search_users, list_pull_requests, list_discussions.
• Context-heavy tools: get_file_contents (scales with file size), list_workflows (fixed ~3.5K regardless of request), list_label when called without a known label name.

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Generated by 🔍 GitHub MCP Structural Analysis · agent · 140 AIC · ⌖ 16.5 AIC · ⊞ 11.6K · ◷

• expires on Jul 31, 2026, 4:36 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 #49334.