[mcp-analysis] MCP Structural Analysis - 2026-05-13

[github-actions[bot]]

Analyzed 10 representative GitHub MCP tools across 9 toolsets for response size, schema structure, and usefulness for agentic work. Today's run consumed ~23.6K tokens with an average usefulness rating of 3.8/5. Best-in-class: list_pull_requests, list_discussions, search_repositories (5/5 — clean schemas, complete data). Biggest concern: get_file_contents returns the entire file body (~15.5K tokens for this repo's README) — context-heavy for large files.

Key Findings

• One tool unavailable: get_me returned 403 — current integration cannot access user context
• Context risk concentration: repos toolset alone accounted for >65% of total token usage (driven by full-file responses)
• Pagination patterns differ: list_discussions uses GraphQL-style cursor pagination (excellent), while list_label returns all labels with no pagination control (530 labels for this repo)
• Minimal-output flag works: search_repositories with minimal_output=true gives a flat 150-token response with everything an agent needs

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	10
Toolsets Covered	9
Total Tokens (Today)	23,558
Average Usefulness Rating	3.8/5
Best Rated Tools	list_pull_requests, list_discussions, search_repositories — 5/5
Worst Rated Tool	get_me — 1/5 (403 error)

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
list_pull_requests	pull_requests	⭐⭐⭐⭐⭐	Excellent — embedded head/base/user, all PR essentials
list_discussions	discussions	⭐⭐⭐⭐⭐	Cursor pagination + totalCount; perfect for traversal
search_repositories	search	⭐⭐⭐⭐⭐	minimal_output=true gives flat, focused schema
get_file_contents	repos	⭐⭐⭐⭐	Complete content + SHA, but no pagination — heavy for large files
list_issues	issues	⭐⭐⭐⭐	Rich data including full body; bodies can be very long
list_workflows	actions	⭐⭐⭐⭐	All workflow metadata; wrapped in content[].text JSON string
list_label	labels	⭐⭐⭐⭐	Clean schema but no pagination (530 labels in one call here)
list_code_scanning_alerts	code_security	⭐⭐⭐	Empty result — schema could not be evaluated
list_tags	repos	⭐⭐⭐	Only name + sha; agent must follow up for date/commit metadata
get_me	context	⭐	403 — not accessible by this integration

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	error	1	error_message	403 response
get_file_contents	resource	3	mimeType, text, uri, sha	Wrapped resource with embedded text
list_issues	array	3	number, title, state, labels, user, body	Embedded user object
list_pull_requests	array	4	number, title, head, base, user, assignees, draft, merged	Deeply nested but well-organized
list_workflows	object	2	id, name, path, state, created_at, html_url	JSON serialized inside content[].text
list_code_scanning_alerts	array	1	(empty)	No data returned
list_discussions	object	3	discussions[], pageInfo, totalCount	GraphQL cursor pagination
list_label	object	2	labels[], totalCount, name, color, id, description	Flat, easy to parse
search_repositories	array	1	full_name, description, stargazers_count, topics, language	Minimal flat schema
list_tags	array	2	name, sha	Sparse — only 2 fields per tag

Response Size Analysis

Toolset	Avg Tokens	Tools Tested
repos	7,760	get_file_contents, list_tags
actions	2,625	list_workflows
labels	2,625	list_label
issues	2,250	list_issues
pull_requests	175	list_pull_requests
discussions	175	list_discussions
search	150	search_repositories
context	30	get_me
code_security	8	list_code_scanning_alerts

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Rating	Notes
get_file_contents	repos	15,500	resource	4	Full file content with SHA — heavy for large files
list_workflows	actions	2,625	object	4	30+ workflows returned with full metadata
list_label	labels	2,625	object	4	All 530 labels in one call — no perPage param
list_issues	issues	2,250	array	4	Includes full issue body which can be very long
list_pull_requests	pull_requests	175	array	5	Clean, complete, immediately actionable
list_discussions	discussions	175	object	5	Cursor pagination + totalCount
search_repositories	search	150	array	5	Minimal output flag works well
get_me	context	30	error	1	403 — integration not authorized
list_tags	repos	20	array	3	Only name and sha — need follow-up call
list_code_scanning_alerts	code_security	8	array	3	Empty — could not evaluate schema

30-Day Trend Summary

Metric	Value
Data Points	10 (first run — baseline)
Average Daily Tokens	23,558
Average Rating Trend	baseline established

This is the first run — future runs will populate the trend window.

Recommendations

Based on the analysis:

• High-value tools (rating 5, low context): list_pull_requests, list_discussions, search_repositories — agents should prefer these when discoverable
• Tools needing improvement: get_me (403 blocker), list_tags (sparse), list_label (missing pagination)
• Context-efficient tools (low tokens, high rating): list_pull_requests (175 tok), list_discussions (175 tok), search_repositories (150 tok)
• Context-heavy tools (use sparingly or with filters): get_file_contents (15.5K tok for this README), list_workflows (2.6K tok), list_label (2.6K tok)
• For agent workflows: prefer list_pull_requests over search_pull_requests when possible — the list endpoint returns the same depth of data with simpler arguments
• For file reads: when only metadata is needed, prefer get_repository_tree over get_file_contents to avoid loading full file bodies

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References:

• §25799237542

Generated by GitHub MCP Structural Analysis · ● 12.7M · ◷

• expires on May 14, 2026, 12:44 PM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-05-14T12:44:15.420Z.

Closed by Workflow