MCP Server

Use AI Atlas NYC inside your AI assistant.

Query the public market map as structured data from Claude, Codex, and other MCP-compatible tools. The server is read-only, local-first, and scoped to published AI Atlas data.

Run locally View tools

Mode

Read-only

Scope

Public

Tools

7 tools

Tools

Structured access to the map.

All tools return concise JSON and canonical AI Atlas URLs. Limits are capped at 25 results.

search_companies

Search published companies by query, category, stage, hook, description, slug, or founder.

get_company

Return one public company profile by slug, including Atlas URL, website, category, stage, founders, funding fields when public, and related patterns.

list_categories

List category metadata, thesis notes, company counts, and optional representative companies.

get_category

Return one category with its public published companies.

list_patterns

List editorial market patterns with framing and related companies.

get_pattern

Return one pattern with related public companies and one-liners.

search_jobs

Search public open jobs when available, returning role metadata and source URLs only.

What it exposes

Published company profiles, category metadata, editorial patterns, public profile URLs, website URLs, public funding fields when present, and public open jobs when available.

What stays private

Admin company management data
Auth profiles, sessions, or private user data
Submissions or submitter contact details
Raw source records, job payloads, agent runs, and automation logs
Supabase keys, service-role keys, cron secrets, or agent write secrets
Ask Atlas LLM calls or private prompt context

Run locally

Connect Claude, Codex, or another MCP client.

The server uses stdio transport and needs no new secrets. In MCP client configs, use silent npm mode so stdout remains valid JSON-RPC.

npm run mcp:server

Claude Desktop

{
  "mcpServers": {
    "ai-atlas-nyc": {
      "command": "npm",
      "args": [
        "--prefix",
        "/path/to/ai-atlas-nyc",
        "run",
        "--silent",
        "mcp:server"
      ]
    }
  }
}

Codex

[mcp_servers.ai-atlas-nyc]
command = "npm"
args = [
  "--prefix",
  "/path/to/ai-atlas-nyc",
  "run",
  "--silent",
  "mcp:server"
]

Try asking

Prompts that work well.

Find NYC AI companies building for finance teams.

Compare companies in Agent Infrastructure and Model Tools.

Show the patterns involving Hebbia.

Which AI Atlas categories have the clearest enterprise buyer pull?

Prefer a browser surface? Ask Atlas answers natural-language questions over the same public map.

Open Ask Atlas