MCP Integration — Lytmus Docs

Lytmus exposes a Model Context Protocol (MCP) server that lets any MCP-compatible AI tool — Claude Desktop, Cursor, Claude Code, or others — read and write to your job-search data.

This is bidirectional. You can log applications, capture interview questions, and update round outcomes entirely from your AI tool of choice. Lytmus acts as the structured data layer; your AI is the interface.

Generate an API key

Open Settings → MCP / API Keys in Lytmus.
Click New key, give it a name, and choose a scope:
- Read + Write — lets your AI tool log applications, rounds, and questions on your behalf (recommended for daily use).
- Read only — lets your AI tool query your data but not modify it.
Copy the key immediately — it is shown only once.

Your key starts with lyt_live_. Store it in a password manager or your shell profile — never commit it to git.

Connect Claude Desktop

Add the following to your Claude Desktop config file:

Path: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "lytmus": {
      "type": "http",
      "url": "https://lytmus.app/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_KEY_HERE"
      }
    }
  }
}

Replace YOUR_KEY_HERE with your key. Restart Claude Desktop after saving.

Once connected, Claude loads lytmus://upcoming into context automatically at the start of each conversation — no prompting needed.

Connect Cursor

Add the following to .cursor/mcp.json (project-level) or your global Cursor MCP settings:

{
  "mcpServers": {
    "lytmus": {
      "url": "https://lytmus.app/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_KEY_HERE"
      }
    }
  }
}

Connect Claude Code

Run this once in your terminal (or from any Claude Code session using !):

claude mcp add --transport http lytmus https://lytmus.app/api/mcp --header "Authorization: Bearer YOUR_KEY_HERE"

Read tools

Your AI can query your job-search data with these tools:

Tool	What it does
`list_applications`	List applications, filter by status / company / season / archived
`get_application`	Full details of one application including its rounds
`list_rounds`	Interview rounds, filterable by application / upcoming days / outcome
`get_round`	One round with its application context and logged questions
`get_upcoming_interviews`	Scheduled rounds in the next N days (default: 7)
`list_questions`	Captured questions, filter by round / tags / confidence range
`search_questions`	Full-text search across question prompts and solutions
`get_metrics`	Funnel counts, response rate, avg confidence, weakest topic tags

Write tools

With a Read + Write key, your AI can also log and update data:

Tool	What it does
`log_application`	Create a new application (auto-creates company / role / level if needed)
`update_application_status`	Move an application to a new pipeline status
`log_round`	Add an interview round to an application
`update_round`	Set outcome, reschedule, or update interviewers
`log_question`	Capture a question with tags, confidence, difficulty, and outcome
`update_question`	Edit a question's text, approach, solution, confidence, or tags
`add_note`	Attach a note to an application, round, or question

Resources

Resources are loaded into your AI's context automatically. Your AI can also request them explicitly by URI.

URI	What it surfaces
`lytmus://upcoming`	Your scheduled interviews in the next 7 days
`lytmus://active`	Applications currently in the interviewing stage with round counts
`lytmus://digest`	Job-search digest: funnel, response rate, avg confidence, weakest tags

Prompts

Prompts are pre-built templates your AI can invoke by name. Useful for consistent prep and debrief workflows.

Prompt	What it does
`interview_prep`	Focused prep brief for an upcoming round — requires `round_id`
`post_interview_debrief`	Structured debrief after finishing a round — requires `round_id`
`weekly_review`	Weekly job-search health check across all active applications

To use a prompt: ask your AI "Use the interview_prep prompt for round <id>", or simply describe what you want and the AI will invoke the right one.

Worked examples

Log an interview right after it ends

"I just finished my Google SRE loop — senior level. I had a system design round on distributed caching, I'd give myself a 2/5. And a behavioral round on dealing with ambiguity, more like a 4/5."

Your AI will call log_application (if Google isn't already logged), log_round twice, and log_question twice — capturing everything while the details are fresh.

Prep for tomorrow's interview

"What interviews do I have tomorrow? What should I focus on?"

Your AI calls get_upcoming_interviews, then get_round for the details, then search_questions to surface similar questions from your past rounds. It'll tell you which topics came up before and how confident you've been on them.

Get a quick job-search briefing

"How's my job search going?"

Your AI reads lytmus://digest and calls get_metrics to give you a summary: applications in flight, response rate, which topics are dragging your confidence, and what's coming up next.

Best practices

Use Read + Write for active searching. A read+write key lets your AI log rounds and questions right after each interview — while the details are still fresh — without ever opening the Lytmus app.

Use Read only for passive use. If you just want prep summaries or progress briefings, a read-only key limits blast radius in case the key is ever exposed.

Start sessions with a digest. Ask your AI to read lytmus://digest at the start of a prep session for a quick briefing on where you stand.

One key per client. Give each client (Claude Desktop, Cursor, Claude Code) its own named key. That way you can revoke one without affecting the others.

Rotate keys when you change machines. Revoke old keys from Settings → MCP / API Keys.

Troubleshooting

401 Unauthorized — The key is missing, wrong, or revoked. Generate a new key in Settings.

Tool not appearing in your AI client — Restart the client fully after adding the config. Some clients need a complete app restart, not just a reload.

interview_prep says round not found — The prompt needs a valid round_id UUID. Call get_upcoming_interviews first to get IDs for your scheduled rounds.

Key accidentally committed to git — Revoke it immediately in Settings → MCP / API Keys and generate a new one. The old key is invalidated the moment you revoke it.