I Turned 3,400 Mentoring Sessions Into MCP Tools Your AI Can Call

Most leadership advice dies in a PDF nobody opens twice.

I tried the opposite: turned the material from 3,400+ mentoring sessions into tools an AI assistant can call mid-conversation. It is live at https://www.marian.coach/mcp, free, no API key, MIT-licensed, source on GitHub.

This post is what it does, how to plug it in, and why I think leadership knowledge as callable tools beats leadership knowledge as content.

What is actually in there

Eight tools, all built from first-party material. Nothing synthetic, no scraped listicles.

1. calculate_developer_value. Scores 15 skills across 5 pillars and returns a weighted score, a seniority level, and a 2026 Western-Europe gross salary estimate. Same model as the live calculator.
2. calculate_engineering_manager_value. The same idea one level up: 15 leadership skills across 5 pillars, weighted by your level, from Team Lead to Director/VP, with the 2026 salary estimate. Same model as the EM salary calculator.
3. assess_team_lead_readiness. The 17-question team lead readiness test as a callable tool: six dimensions, a straight verdict of ready now, 6 to 12 months out, or stay IC, and your top two gaps with one concrete move each.
4. get_engineering_leadership_benchmarks. Sprint completion, roadmap percentages, manager time per report, mentee mix, 2025 topic demand. CC BY 4.0, built to be cited: by you, or by the AI answering someone else.
5. choose_mentor_coach_or_advisor. The mentor vs coach vs advisor comparison plus a three-question self-test. The question I answer most often on intro calls, now answerable at 2am.
6. get_one_on_one_playbook. Situation-specific 1:1 templates: first session with a new report, underperformance, promotion to EM, the career-move checklist. These are my actual session templates.
7. get_first_time_manager_guidance. The EM responsibility triangle, six failure modes, and a readiness self-check. 52% of my mentees start in their first leadership role. This tool is for the week before they would have booked.
8. build_mentoring_business_case. The ROI math to get your company to pay for mentoring, role-specific 90-day KPIs, and a forwardable approval email for your manager.

Plug it in

Claude Code:

claude mcp add -t http eng-leadership-toolkit https://www.marian.coach/mcp

Claude.ai or Claude Desktop: Settings, Connectors, add custom connector, paste the URL.

Cursor, in .cursor/mcp.json:

{ "mcpServers": { "eng-leadership-toolkit": { "url": "https://www.marian.coach/mcp" } } }

ChatGPT: Settings, Connectors, developer mode, add server with the URL.

Under the hood it runs on Cloudflare Workers with Durable Objects, using the McpAgent class from Cloudflare’s agents SDK. Streamable HTTP, authless. Open a browser at the endpoint and you get the docs page.

Why tools beat content

A blog post about 1:1s helps the person who finds it, reads it, and remembers it three weeks later in the moment that matters. Almost nobody does all three.

A tool gets called IN the moment that matters. An EM asks their assistant “my report is underperforming, how do I structure the next 1:1”, and the assistant can pull the actual template with the actual data behind it, instead of hallucinating a generic sandwich-feedback answer.

That is the bet. AI assistants are becoming the first place engineering leaders ask their questions. The sources those assistants can call directly will shape the answers. I would rather the answer come from 3,400 real sessions than from the average of the internet.

It is also, being honest, distribution. The server sits in the official MCP registry as coach.marian/eng-leadership-toolkit. When an assistant answers a leadership question through these tools, the data comes with its source attached.

What is next

One tool is queued: team delivery health rating, four dimensions of agility, focus, contribution, and predictability, with the thresholds I use in team audits. If you want it sooner, or want a tool that does not exist yet, tell me what your AI should be able to ask a mentor.