[mcp-analysis] MCP Structural Analysis - 2026-03-04

[github-actions[bot]]

Today's run analyzed 9 GitHub MCP tools across 9 toolsets. Average usefulness rating was 2.78/5 — driven down by tools that error (context, code_security) and tools with oversized payloads (actions, pull_requests). Best performer: search_repositories (5/5). Worst: get_me & list_code_scanning_alerts (1/5 each, both error responses).

Full Structural Analysis Report

Executive Summary

Metric	Value
Tools Analyzed	9
Total Tokens (Today)	11,754
Average Usefulness Rating	2.78/5
Best Rated Tool	search_repositories: 5/5
Worst Rated Tool	get_me, list_code_scanning_alerts: 1/5

---

Usefulness Ratings for Agentic Work

Tool	Toolset	Rating	Assessment
search_repositories	search	⭐⭐⭐⭐⭐	Excellent — compact, structured, all key metadata
list_issues	issues	⭐⭐⭐⭐	Good — rich data but body text inflates tokens
list_discussions	discussions	⭐⭐⭐⭐	Good — clean list, no body bloat, category embedded
list_label	labels	⭐⭐⭐⭐	Good — flat structure, 26 labels returned efficiently
get_file_contents	repos	⭐⭐	Limited — only returns SHA, no file content surfaced
list_pull_requests	pull_requests	⭐⭐	Limited — offloaded to file (18,784 chars); too large
list_workflows	actions	⭐⭐	Limited — offloaded to file (16,833 chars); 94 workflows bloat
get_me	context	⭐	Poor — 403 error, returns no data for integration token
list_code_scanning_alerts	code_security	⭐	Poor — 404, no scanning configured for this repo

---

Schema Analysis

Tool	Type	Depth	Key Fields	Notes
get_me	error	0	—	Integration token 403; no data
get_file_contents	object	1	sha	Only SHA returned; content not in MCP response
list_issues	array	3	number, title, state, body, labels, user	Full body included; paginated
list_pull_requests	array	4	id, number, state, title, body, user, head, base	Deeply nested; payload file offloaded
list_workflows	object	2	total_count, workflows[id,name,path,state]	94 workflows; file offloaded
list_code_scanning_alerts	error	0	—	404; no code scanning on repo
list_discussions	array	2	number, title, category, html_url, user	Ideal list format; no body bloat
list_label	array	1	name, color, description, id	Flat, complete, efficient
search_repositories	object	2	total_count, items[id,name,description,language,stats]	Best balance of completeness + size

---

Response Size Analysis

Toolset	Avg Tokens	Notes
pull_requests	4,696	File offload; bloated even per-PR
actions	4,208	File offload; 94 workflows ignores perPage
issues	1,750	Body text is primary driver
labels	750	26 labels, reasonable
search	150	Best efficiency
discussions	125	Minimal, well-structured
repos	25	Minimal (SHA only)
context	25	Error only
code_security	25	Error only

---

Tool-by-Tool Analysis

Tool	Toolset	Tokens	Schema	Depth	Rating	Notes
get_me	context	25	error	0	1	403 — integration token restriction
get_file_contents	repos	25	object	1	2	SHA-only; actual content not visible
list_issues	issues	1,750	array	3	4	Rich but body inflates size
list_pull_requests	pull_requests	4,696	array	4	2	File offload; high cost per PR
list_workflows	actions	4,208	object	2	2	File offload; ignores perPage
list_code_scanning_alerts	code_security	25	error	0	1	404 — no scanning configured
list_discussions	discussions	125	array	2	4	Clean, efficient list format
list_label	labels	750	array	1	4	Flat + complete
search_repositories	search	150	object	2	5	Best ratio: low tokens, high value

---

30-Day Trend Summary

Metric	Value
Data Points	36 (4 run dates)
Dates Covered	2026-02-27 → 2026-03-04
Average Daily Tokens	~11,300
Average Rating (all runs)	~2.78/5 (stable)
Trend	Stable — consistent toolset behavior

---

Recommendations

High-value tools (rating 4-5):

• search_repositories — use as primary repo discovery tool
• list_discussions, list_label, list_issues — well-structured, actionable

Tools needing improvement:

• list_workflows / list_pull_requests — need server-side pagination enforcement; perPage ignored
• get_file_contents — should surface decoded file content in MCP response, not just SHA
• get_me — should work for integration (GitHub App) tokens, not just PATs

Context-efficient tools (low tokens, high rating):

• search_repositories (150 tokens, 5/5) — ideal for agents
• list_discussions (125 tokens, 4/5) — great list format

Context-heavy tools (avoid or filter output):

• list_pull_requests — consider filtering fields server-side
• list_workflows — truncate or enforce perPage at MCP layer

---

Visualizations

Response Size by Toolset

Usefulness Ratings

Daily Token Trend

Size vs Usefulness

References: §22667668100

AI generated by GitHub MCP Structural Analysis

• expires on Mar 5, 2026, 11:42 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-03-05T11:42:39.249Z.

Closed by Workflow