v0.2.0: ask your network from Claude and other AI tools

Network Constellation v0.2.0 — ask your network from Claude and other AI tools

Your LinkedIn network, available to your AI assistant. Ask Claude "who do I know in satellite imagery?", "who do I know at Esri?" or "which three VCs do I know best?", and it answers from your own Connections.csv.

Claude Desktop: one click

Download network-constellation.mcpb below, double-click it, choose your Connections.csv (not the .mcpb itself), and start a new chat. You don't need Node, git or any config.

Claude Code, Cursor, VS Code, Windsurf, Gemini CLI, Codex

claude mcp add network-constellation --scope user -- npx -y github:asturksever/network-constellation --csv /full/path/to/Connections.csv

Setup for every client: docs/mcp.md.

Six read-only tools

network_overview, network_ask, network_find_person, network_list_people, network_list_employers, network_get_employer. They use the same classifier and question engine as the web app, and never connect to the internet.

What your AI tool sees

The file stays on your machine, but the answers to your questions (names, job titles and employers of the people who match) become part of your conversation, so they go to your AI provider. Email addresses are never returned. Profile links are returned only if you turn them on.

Still no dependencies: the server speaks MCP directly and runs straight from GitHub.

v0.1.0 — first public release

Network Constellation v0.1.0 — first public release

See who you know. Drop in your LinkedIn connections and they arrange themselves
in 3D by what people do: thousands of names become a few dozen clusters you can
orbit, search and question in plain English. It runs entirely in your browser,
and nothing is uploaded.

Try it: https://asturksever.github.io/network-constellation/ (demo included:
250 invented people)

What's in it

• A landing page you can try cold. The demo turns behind the page. Explore
  it, or run an example question, or drop your own Connections.csv.
• The constellation. Everyone is placed by field and joined to shared
  employers. Orbit, isolate a field, colour by seniority, or find anyone by
  name.
• Ask in plain English. "who can intro me to a VC?" is answered locally
  in milliseconds, with no model and no key. The query it ran is shown above the
  answer, and the matches light up in the scene.
• Profiles. Click anyone for their headline, employer and everyone else
  you know there.
• Bring your own key (optional). With it, Claude places employers (so
  "any VC based in London?" works) and reads questions more closely. It also
  researches one person on the public web, with sources and a confidence
  rating, but only when you press Enrich profile.
• Works on a phone, as bottom sheets.
• Single-file builds. npm run bundle:demo makes one HTML file you can
  hand to anyone.

Privacy, in one paragraph

Your file is read in your tab and kept in your browser. Without a key, the
only other request the page makes is for the 3D library, from jsDelivr. With
your key, three things can reach Anthropic, and nothing else: employer names,
your question text, and one person's name, headline and employer when you
press Enrich profile. Never emails or profile links.
Details.

Docs

Tutorial ·
Install & self-host ·
How questions are answered

MIT licensed. No dependencies. Contributions to src/taxonomy.js are the most
useful of all: if your field lands in "Other", that's the file to fix.