[mcp-analysis] MCP Structural Analysis - 2026-08-04

[github-actions[bot]]

MCP Structural Analysis — 2026-08-04

First day of tracking. 10 representative calls sampled across 8 toolsets plus the injected workflow-context block. Average usefulness: 3.1/5. Two calls failed outright due to external constraints (a search API secondary rate-limit, and a missing permission grant) rather than schema problems — these are noted separately from genuine design weaknesses below.

Metric	Value
Tools Analyzed	10
Total Tokens (Today)	~9,875
Average Usefulness Rating	3.1/5
Best Rated Tools	github_context (workflow-injected) & list_pull_requests: 5/5
Worst Rated Tools	get_teams & search_code: 1/5 (both failed for external reasons, not schema quality)

Full Structural Analysis Report

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
github_context	workflow_context	⭐⭐⭐⭐⭐	Excellent — free, always-available identity data
list_pull_requests	pull_requests	⭐⭐⭐⭐⭐	Excellent — compact, honored perPage, immediately actionable fields
list_code_scanning_alerts	code_security	⭐⭐⭐⭐	Good — clean empty-array response, efficient
list_discussions	discussions	⭐⭐⭐⭐	Good — compact, includes pagination cursor + related entities
list_issues	issues	⭐⭐⭐	Adequate — perPage honored, but sole result was fully redacted by an integrity-policy filter this run
list_workflows	actions	⭐⭐⭐	Adequate — rich fields, but ignored per_page=1 and returned 30 of 347 workflows
list_label	labels	⭐⭐⭐	Adequate — no pagination/filter controls, always returns the full label set
get_file_contents	repos	⭐⭐	Limited — no excerpt/summary option; this README response was dominated by a large inline contributor-attribution list
get_teams	context	⭐	Poor this run — blocked by a missing permission grant, not a schema issue
search_code	search	⭐	Poor this run — failed with a 429 secondary rate limit from GitHub's search API

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
github_context	markdown	1	actor, repository, workspace, workflow-run-id	Plain bullet list, zero MCP overhead
list_pull_requests	array	1	number, title, state, html_url	Flat, no nesting
list_code_scanning_alerts	array	1	(none — empty)	—
list_discussions	object	3	discussions[], pageInfo, totalCount	category & user embedded per discussion
list_issues	array	1	(none — redacted)	Integrity filter hid the only item
list_workflows	object	2	total_count, workflows[].{id,name,path,state,html_url}	—
list_label	object	2	labels[].{id,name,color,description}, totalCount	—
get_file_contents	object	2	content[].resource.text, resource.uri	Raw file text, no metadata-only mode
get_teams	n/a	0	—	Permission-blocked
search_code	error	1	error_message	Rate-limited

Response Size Analysis

Toolset	Avg Tokens	Tools Tested
repos	~4,800	1
labels	~2,500	1
actions	~2,200	1
workflow_context	80	1
discussions	~130	1
issues	~60	1
search	~50	1
pull_requests	~40	1
code_security	~15	1
context	0	1 (blocked)

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
github_context	workflow_context	80	markdown, depth 1	5	Free identity block
get_teams	context	0	n/a	1	Permission not granted for this call
get_file_contents	repos	~4,800	object, depth 2	2	README bloated by contributor list; no partial-read option
list_issues	issues	~60	array, depth 1	3	Only result redacted by integrity policy
list_pull_requests	pull_requests	~40	array, depth 1	5	Clean, exactly matched perPage
list_workflows	actions	~2,200	object, depth 2	3	per_page ignored — got 30 of 347
list_code_scanning_alerts	code_security	~15	array, depth 1	4	Efficient empty result
list_discussions	discussions	~130	object, depth 3	4	Pagination + related entities present
list_label	labels	~2,500	object, depth 2	3	No filter/pagination controls
search_code	search	~50	error	1	429 secondary rate limit

30-Day Trend Summary

Metric	Value
Data Points	10 (day 1 of tracking)
Average Daily Tokens	~9,875
Average Rating Trend	Baseline established — no prior data to compare

Recommendations

• High-value tools (rating 4–5): github_context (free), list_pull_requests, list_discussions, list_code_scanning_alerts — prefer these when the equivalent data is needed.
• Tools needing improvement: list_workflows and list_label ignore their pagination parameters (per_page/perPage had no visible effect), forcing agents to consume more context than requested — worth filing upstream. get_file_contents offers no way to fetch a summary or byte-range of a large file.
• Context-efficient + high-value: list_pull_requests, list_discussions, list_code_scanning_alerts — small payloads, high signal.
• Context-heavy tools: get_file_contents (~4,800 tokens for one README) and list_label (~2,500 tokens for a label lookup) — use targeted alternatives (e.g., get_label for a single label) where possible.
• Note on this run's failures: get_teams (permission not granted) and search_code (GitHub API 429) failed for reasons external to the MCP schema itself — re-test these once the constraint clears rather than treating them as permanent low scores.
• Note on filtering: list_issues returned an item that was fully redacted by an "integrity policy" before reaching the agent, with only a generic reason string — worth understanding if this is expected behavior for this repository's issue content.

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

Generated by 🔍 GitHub MCP Structural Analysis · agent · 125 AIC · ⌖ 40.6 AIC · ⊞ 11.6K · ◷

• expires on Aug 5, 2026, 4:42 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 #50551.