seco

a local professional identity engine. one intake session, every surface.

what it does

seco coaches you through a voice conversation about your work, stores your experiences locally, and renders optimized copy for wherever you need to show up professionally — resume, LinkedIn, GitHub, Overleaf, cover letters.

your data stays on your machine. you bring your own API keys.

install

npx seco

on first run, seco walks you through key setup and prints the one-line config to add to Claude Desktop.

requirements

• Node.js 20+
• Anthropic API key — required
• Deepgram API key — optional, for real-time voice intake
• OpenAI API key — optional, Whisper fallback for voice

claude desktop setup

after running npx seco once, add this to your Claude Desktop config:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "seco": {
      "command": "npx",
      "args": ["seco"]
    }
  }
}

restart Claude Desktop. you can now say things like:

"start a new intake session"
"render my last three experiences as resume bullets"
"tailor everything to this job description: [paste JD]"
"export my internship experience as LaTeX"

voice intake

for the voice-coached intake flow, open http://localhost:3000 after starting seco. this requires a Deepgram key for real-time transcription, or an OpenAI key for the Whisper fallback.

output surfaces

| surface | format | |---|---| | resume bullets | ATS-optimized, action verb + metric | | linkedin summary | ~300 words, first-person prose | | linkedin post | hook-first, 150–300 words | | github readme | markdown, technical framing | | overleaf / latex | \item blocks, paste-ready | | cover letter paragraph | modular, one experience per paragraph | | bio (short / medium / full) | third-person, variable length |

mcp tools

| tool | description | |---|---| | start_intake_session | begin a voice or text coaching session | | save_experience | commit session to local database | | list_experiences | browse entries with optional filters | | get_experience | retrieve a single entry | | render_for_surface | generate copy for any surface | | tailor_to_jd | full tailoring pipeline against a job description | | export_latex | emit LaTeX bullet blocks | | update_experience | targeted field edit | | delete_experience | remove an entry |

data

experiences are stored in a local SQLite database at ~/.seco/seco.db. nothing is sent to any server except your own API calls to Anthropic and Deepgram.

optional cross-device sync via Supabase — set SUPABASE_URL and SUPABASE_ANON_KEY in ~/.seco/.env to enable.

contributing

see AGENTS.md for the architectural rules before opening a PR.

the most welcome contribution is a new render surface — add a prompt builder in packages/core/src/render/prompts/ and register it in the surface map. no other files need to change.

license

MIT