[mcp-analysis] MCP Structural Analysis - 2026-08-06

[github-actions[bot]]

Warning

Threat Detection Engine Failure — The analysis engine could not complete. This is a tooling failure, not a security finding.

What happened

The threat detection engine failed to produce results.

Review the workflow run logs for details.

Overview

Baseline day for the 30-day MCP structural trend (this is the first data point — no prior history in cache-memory). Analyzed 10 tool responses across 9 toolsets plus the injected workflow-context block. Average usefulness: 3.3/5. Best: github_context (5/5, zero MCP overhead). Worst: get_teams (1/5 — blocked by a permission gate this run, returned no data).

Key metrics

Metric	Value
Tools analyzed	10
Total tokens (today)	~5,835
Average usefulness rating	3.30 / 5
Best rated	github_context: 5/5
Worst rated	get_teams: 1/5 (permission-blocked)

Notable findings

• get_teams was blocked by permission, not by data absence — the tool required a grant this session didn't have, so the context toolset returned zero usable data. Worth checking whether the read-only design-scope docs match actual runtime gating.
• list_issues returned an empty array because the one candidate issue was filtered out by content-integrity policy ("lower integrity than agent requires"), not because no issues existed. Agents relying on list_issues counts could undercount silently.
• get_file_contents on README.md was 60%+ padding: a ~7,000-character flat list of community-contributor links diluted a ~9,500-character response. No section/range read was available for this call.
• list_label has no perPage support in the built-in tool and returned ~100 of 843 labels in one uncontrolled shot — a repo with a large label taxonomy (like this one) risks oversized responses.
• The mcpscripts list_workflows wrapper correctly honored perPage, unlike the built-in GitHub MCP list_workflows, which silently ignores camelCase perPage.
• list_code_scanning_alerts carried actionable file/line locations, but each alert embedded a long rule.help markdown block (examples, CWE references) that roughly tripled the payload size versus a summary-only response.

Full Structural Analysis Report

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
github_context (injected)	workflow_context	⭐⭐⭐⭐⭐	Excellent — zero-cost, immediately actionable identity data
list_pull_requests	pull_requests	⭐⭐⭐⭐	Good — compact, fields filter worked well
list_workflows (mcpscripts)	actions	⭐⭐⭐⭐	Good — correct pagination, clean schema
list_discussions	discussions	⭐⭐⭐⭐	Good — cursor pagination + totalCount upfront
search_code	search	⭐⭐⭐⭐	Good — minimal, total_count useful for existence checks
get_file_contents	repos	⭐⭐⭐	Adequate — useful content but diluted by a long link list
list_code_scanning_alerts	code_security	⭐⭐⭐	Adequate — actionable but bloated by embedded help text
get_label	labels	⭐⭐⭐	Adequate — no perPage control, risk of oversized responses at scale
list_issues	issues	⭐⭐	Limited — integrity-policy filtering silently zeroed the result
get_teams	context	⭐	Poor — blocked by an ungranted permission, no data returned

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
github_context	markdown	1	actor, repository, workspace, workflow-run-id	Prompt-injected, no MCP round-trip
get_teams	error	0	—	Permission-gated in this run
get_file_contents	object (resource)	3	content[].resource.text/uri/mimeType	Whole-file fetch, resource envelope adds nesting
list_issues	array	1	number, title, state, labels	Empty due to integrity filter
list_pull_requests	array	1	number, title, state, html_url	Clean, flat
list_workflows	object	2	total_count, per_page, page, workflows[]	Pagination metadata present
list_code_scanning_alerts	array	3	number, rule.id/severity/help, most_recent_instance.location	Deep nesting via rule + instance objects
list_discussions	object	3	discussions[].category/title/user, pageInfo, totalCount	Cursor-based pagination
get_label	object	2	labels[].name/color/description/id, totalCount	No perPage control
search_code	object	2	items[].name/path, total_count, incomplete_results	Compact search envelope

Response Size Analysis

Toolset	Avg Tokens	Tools Tested
repos	~2,375	1
code_security	~1,500	1
labels	~1,500	1
actions	~125	1
discussions	~100	1
workflow_context	~75	1 (prompt-injected)
issues	~65	1
search	~40	1
pull_requests	~30	1
context	~25	1 (error)

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	75	markdown	⭐⭐⭐⭐⭐	No MCP call required
list_pull_requests	pull_requests	30	array	⭐⭐⭐⭐	fields filter worked well
list_workflows	actions	125	object	⭐⭐⭐⭐	mcpscripts wrapper honors perPage correctly
list_discussions	discussions	100	object	⭐⭐⭐⭐	Cursor pageInfo + totalCount
search_code	search	40	object	⭐⭐⭐⭐	Minimal, fields filter effective
get_file_contents	repos	2375	object (resource)	⭐⭐⭐	~70% of content was a low-value contributor list
list_code_scanning_alerts	code_security	1500	array	⭐⭐⭐	Long embedded rule.help text triples payload
get_label	labels	1500	object	⭐⭐⭐	No perPage; 843 total labels in this repo
list_issues	issues	65	array	⭐⭐	Integrity policy silently filtered the only candidate
get_teams	context	25	error	⭐	Blocked by ungranted permission

30-Day Trend Summary

Metric	Value
Data Points	1 (baseline — first run)
Average Daily Tokens	5,835
Average Rating Trend	Not yet available (need ≥2 days)

Recommendations

• High-value tools (rating 4-5): github_context (injected), list_pull_requests, list_workflows (mcpscripts wrapper), list_discussions, search_code — prefer these for cheap, decisive reads.
• Tools needing improvement: get_teams (permission gating blocks reads that docs suggest should be prompt-free), list_label (no perPage — should default to a bounded page size at 843 labels in this repo), list_code_scanning_alerts (strip rule.help boilerplate from default responses, or offer a summary mode).
• Context-efficient tools (low tokens, high rating): list_pull_requests, search_code, list_discussions — all under 130 tokens with fields/pagination support.
• Context-heavy tools: get_file_contents (README) and list_code_scanning_alerts/get_label (~1,500 tokens each) — each has a specific, addressable bloat source (link list, rule.help text, unbounded label page).

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Generated by 🔍 GitHub MCP Structural Analysis · agent · 110.2 AIC · ⊞ 11.9K · ◷

• expires on Aug 7, 2026, 4:41 AM UTC-08:00

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-08-07T12:41:43.672Z.

Closed by Workflow