Loading ToolBench

Reading the current corpus…

connectsafely-cookbook MCP server · F · ToolBench

connectsafely-cookbook

github.com/ConnectSafelyAI/connectsafely-cookbook

A collection of example skills and snippets for using the ConnectSafely LinkedIn API, including prospect research and profile fetching utilities.

2026-07-28+STDIO5 tools

42/100

Definition Quality49

Protocol Readiness25

Supportability40

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

This server exposes 5 tools for LinkedIn profile research via a Python skill template. All tools have descriptions (10-80 chars) and basic input schemas with type info. However, descriptions are uniformly terse (well below the 194-char baseline), parameter descriptions are minimal or absent, output schemas are not documented, and error handling is weak. Tools are READ_ONLY but lack idempotent/readOnly annotations. The codebase shows no MCP server registration, only standalone Python scripts calling an external API. Tool definitions appear inferred from example.py rather than explicitly registered in an MCP server context.

Tools (5)

build_briefread onlysource verified47/100

Reduce raw API responses into a structured outreach brief with snapshot, background, recent posts, and conversation hooks

extract_profile_idread onlysource verified50/100

Accept a full LinkedIn URL or a bare profileId and return the profile slug

fetch_latest_postsread onlyauthsource verified52/100

Fetch the latest posts from a LinkedIn profile

fetch_profileread onlyauthsource verified53/100

Fetch a LinkedIn profile by public ID with optional skills and education data

render_markdownread onlysource verified45/100

Render a research brief as formatted Markdown output

Issues (5)

CRITICALtool ↗

No output schemas documented. Tools return objects (profile_payload, brief, markdown) but LLMs cannot infer what fields to expect or how to chain results downstream.

fetch_profilefetch_latest_postsbuild_briefrender_markdown

HIGHtool-description ↗

Parameter descriptions are minimal or missing. 'profile_payload' and 'posts_payload' in build_brief lack any description of expected structure. 'count' in fetch_latest_posts has no guidance on valid range.

build_brieffetch_latest_posts

HIGHtool-description ↗

Tool descriptions are terse (25 - 40 chars vs 194-char baseline). 'Render a research brief as formatted Markdown output' lacks context on when to call it or what the output format is.

render_markdownfetch_latest_posts

MEDIUMtool ↗

No tool annotations (readOnlyHint, idempotentHint). All tools are READ_ONLY and idempotent, but this is not declared in the schema, forcing LLMs to infer safety.

Recommendations

Document output schemas for all tools. E.g., fetch_profile should declare: {type: 'object', properties: {id, name, headline, skills: [...], education: [...]}, required: [id, name]}.
Expand tool descriptions to 100 - 200 chars. Include WHEN to call (e.g., 'Call after fetch_profile to extract actionable insights for outreach') and WHAT it returns.
Add descriptions to all parameters. E.g., 'count: Number of recent posts to fetch (1 - 50, default 10)' instead of just 'Number of posts to fetch'.
Add tool annotations: readOnlyHint=true and idempotentHint=true to all tools so LLMs know they are safe to retry.
Implement an actual MCP server (e.g., using mcp-python or similar) that registers these tools with the MCP protocol, rather than standalone scripts. This enables remote client access and proper tool discovery.

Score history

Overall score trend

First recorded score · v2 rubric

42/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	42	2026-07-28+	v2