[mcp-analysis] MCP Structural Analysis - 2026-02-20

[github-actions[bot]]

Analyzed 11 GitHub MCP tools across 9 toolsets on 2026-02-20. Response sizes range from 13 tokens (get_me error) to 24,713 tokens (list_code_scanning_alerts). Average usefulness rating: 3.27/5. Two tools scored 5/5 (list_discussions, search_repositories); the code security toolset is a major context concern at 98KB per response.

Key Findings

• list_code_scanning_alerts returns 98KB including full rule descriptions and help text — far too large for routine agent use
• list_discussions and search_repositories are the most context-efficient paginated tools
• PR and workflow list tools exceed context limits even at perPage=1 due to body text — results saved to file
• Specific get_* tools (e.g. get_label) are dramatically more efficient than their list_* counterparts
• get_me returns a 403 with the current integration token — context toolset unavailable here

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	11
Toolsets Covered	9
Total Tokens (Today)	39,323
Average Usefulness Rating	3.27 / 5
Best Rated Tools	list_discussions, search_repositories — 5/5
Worst Rated Tool	get_me — 1/5 (403 error)
Most Expensive Tool	list_code_scanning_alerts — 24,713 tokens
Most Efficient Tool	get_label — 23 tokens, rating 4/5

---

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
list_discussions	discussions	⭐⭐⭐⭐⭐	Excellent — compact, paginated, all key fields
search_repositories	search	⭐⭐⭐⭐⭐	Excellent — rich metadata, compact response
list_commits	repos	⭐⭐⭐⭐	Good — clean structure with author details
list_issues	issues	⭐⭐⭐⭐	Good — rich data; body inflation risk
get_label	labels	⭐⭐⭐⭐	Good — efficient single lookup, flat schema
list_pull_requests	pull_requests	⭐⭐⭐	Adequate — too large for context, body inflates
list_workflows	actions	⭐⭐⭐	Adequate — still large even at minimal page size
list_label	labels	⭐⭐⭐	Adequate — fetches all labels regardless of limit
list_code_scanning_alerts	code_security	⭐⭐	Limited — 98KB response; rule help text is severe bloat
get_file_contents	repos	⭐⭐	Limited — returns minimal confirmation, content not surfaced inline
get_me	context	⭐	Poor — 403 Not accessible by integration

---

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	error	0	—	403 forbidden with integration token
get_file_contents	text	0	sha, content	Only returns success confirmation; content opaque
list_commits	array	3	sha, html_url, commit.message, author, committer	Clean nested structure; committer sub-object useful
list_issues	object	4	number, title, state, body, user, labels, comments, pageInfo	labels and user embedded; body text drives size variance
list_pull_requests	array	4	id, number, state, title, body, user, labels, head, base	head/base branches useful; body drives context overflow
list_workflows	object	3	id, name, path, state, url, html_url, badge_url	199 workflows in repo; badge_url and html_url redundant for agents
list_code_scanning_alerts	array	4	number, rule.id, rule.severity, rule.description, state, location	Full help text + CWE tags per alert = extreme bloat
list_discussions	object	3	number, title, category, user, html_url, created_at, pageInfo	Category sub-object lean; pagination metadata clean
list_label	object	2	id, name, color, description	Flat per-label; fetches all regardless of limit param
get_label	object	1	color, description, id, name	Ideal: 4-field flat response; no nesting
search_repositories	object	2	total_count, items.name, items.description, items.language, items.stargazers_count, items.topics	Topics array very useful for discovery

---

Response Size Analysis

Toolset	Total Tokens	Tools Tested
code_security	24,713	1
pull_requests	4,457	1
actions	3,885	1
issues	3,000	1
labels	2,778	2
repos	226	2
search	138	1
discussions	113	1
context	13	1

---

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema Type	Depth	Rating	Notes
get_me	context	13	error	0	1	403 - integration token blocked
get_file_contents	repos	21	text	0	2	SHA confirmation only; content not surfaced
get_label	labels	23	object	1	4	Ideal flat response for targeted queries
list_discussions	discussions	113	object	3	5	Best paginated list tool by efficiency
search_repositories	search	138	object	2	5	Rich + compact; excellent for discovery
list_commits	repos	205	array	3	4	Clean commit history, good for auditing
list_label	labels	2,755	object	2	3	No perPage respect; fetches all labels
list_issues	issues	3,000	object	4	4	Body text dominates; useful when brief
list_workflows	actions	3,885	object	3	3	perPage=1 still returned large payload
list_pull_requests	pull_requests	4,457	array	4	3	Triggered file offload; body + metadata heavy
list_code_scanning_alerts	code_security	24,713	array	4	2	Rule help text is severe and unnecessary bloat

---

30-Day Trend Summary

Metric	Value
Data Points	11 (first run — baseline established)
Average Daily Tokens	39,323
Average Rating Trend	Baseline — no prior data to compare

Historical trending will begin accumulating from this run forward.

---

Recommendations

High-value tools (rating 4-5) — prefer these:

• list_discussions — most efficient paginated list tool
• search_repositories — compact, rich metadata, excellent for discovery
• get_label — use specific lookup over list_label whenever possible
• list_commits — clean, predictable schema

Tools needing improvement:

• list_code_scanning_alerts — should strip rule.help text by default; truncate to id, severity, state, location
• list_pull_requests — should exclude body field by default or offer a fields projection param
• list_workflows — perPage parameter appears to have limited effect; response includes all metadata

Context-efficient sweet spot (compact + useful):

• get_label (23 tokens, rating 4), list_discussions (113 tokens, rating 5), search_repositories (138 tokens, rating 5)

Avoid in context-sensitive workflows:

• list_code_scanning_alerts — 24,713 tokens; use get_code_scanning_alert with specific alert numbers instead
• list_pull_requests — even minimal requests exceed context threshold; prefer targeted pull_request_read with known PR number

---

Visualizations

Response Size by Tool (colored by usefulness)

Usefulness Ratings by Toolset

Token Usage by Toolset

Token Size vs Usefulness Rating

References:

• §22221808944

AI generated by GitHub MCP Structural Analysis

• expires on Feb 27, 2026, 11:17 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-02-27T11:17:46.856Z.

Closed by Workflow