Global Database — MCP Server

Remote MCP server that gives LLM agents access to Global Database: company profiles, financials, ownership, digital insights, people search and contact enrichment, prospecting, and KYB/compliance lookups against official government registries.

Hosted at https://mcp.globaldatabase.com/mcp (Streamable HTTP). Nothing to install or run — you connect to it and sign in with your Global Database API key through the browser.

Install

Cursor

Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

{
  "mcpServers": {
    "global-database": {
      "type": "http",
      "url": "https://mcp.globaldatabase.com/mcp"
    }
  }
}

Claude Code

claude mcp add --transport http global-database https://mcp.globaldatabase.com/mcp

Or install the plugin, which adds the server and three skills (company-due-diligence, company-snapshot, prospect-list):

/plugin marketplace add global-database/mcp-server
/plugin install global-database@global-database

VS Code (GitHub Copilot)

Use the Install server badge above, or add it from the terminal:

code --add-mcp '{"name":"global-database","type":"http","url":"https://mcp.globaldatabase.com/mcp"}'

To commit it for a whole repository instead, put this in .vscode/mcp.json — the copy in this repo is exactly that file, so cloning it and opening the folder in VS Code offers the server without any further setup:

{
  "servers": {
    "global-database": {
      "type": "http",
      "url": "https://mcp.globaldatabase.com/mcp"
    }
  }
}

Note the key is servers, not mcpServers. VS Code differs from Claude Code and Cursor here, the same way Gemini CLI expects httpUrl and Antigravity expects serverUrl for the identical endpoint. Copying the wrong file between clients produces a config that loads without error and connects to nothing.

A user-level equivalent exists too — run MCP: Open User Configuration from the Command Palette (it lives under the VS Code user profile, not ~/.copilot/mcp-config.json — that path belongs to the separate GitHub Copilot CLI and expects mcpServers, not servers).

Signing in is a browser OAuth flow on first use; there is nothing to paste into the config.

Gemini CLI

Install the bundled extension straight from this repo:

gemini extensions install https://github.com/global-database/mcp-server

Or add the server to settings.json by hand:

{
  "mcpServers": {
    "global-database": {
      "httpUrl": "https://mcp.globaldatabase.com/mcp",
      "authProviderType": "dynamic_discovery"
    }
  }
}

Perplexity

Requires a paid plan (custom connectors are a Pro/Enterprise feature). In Settings → Connectors → + Custom connector → Remote, enter:

Then open the connector card to run the OAuth sign-in.

Grok

At grok.com/connectors → New Connector → Custom, enter the MCP server URL https://mcp.globaldatabase.com/mcp and complete the OAuth sign-in. Grok discovers the tools from the live endpoint.

Microsoft Copilot Studio

In your agent → Tools → Add a tool → Model Context Protocol → New MCP server, enter the server URL https://mcp.globaldatabase.com/mcp and pick OAuth 2.0 with dynamic registration / discovery. Leave Scope empty — the server publishes no scopes, and a non-empty one fails the token request.

Copilot Studio ignores the MCP instructions field, so the agent starts without the routing rules every other client receives. Paste copilot-studio-instructions.md into Agent → Overview → Instructions after the connection is created.

Claude.ai / other MCP clients

Add a custom connector pointing at https://mcp.globaldatabase.com/mcp.

Authentication

OAuth 2.1 with PKCE and Dynamic Client Registration — the client registers itself, so there is nothing to configure. On first connect your browser opens a Global Database login page where you paste your API key; the key is held in the access-token claims server-side and is never passed as a tool argument.

Get an API key at globaldatabase.com.

Tools

Resources and prompts

Beyond tools, the server exposes MCP resources so a client can load reference data once per conversation instead of spending a tool call (and credits) on it:

Larger, hierarchical nomenclatures (NACE, ISIC, SIC, industry focus, region) stay behind the get_nomenclature tool and its search parameter.

Prompts shipped with the server: company_research, due_diligence, lead_generation, competitor_analysis, market_research.

Plugins and skill

The same contents ship as a plugin for both Cursor (.cursor-plugin/plugin.json) and Claude Code (.claude-plugin/marketplace.json), so the server can be installed from a marketplace instead of being wired up by hand. Each plugin bundles:

• the remote MCP server above (all tools), and
• three skills: company-due-diligence walks the agent through a structured KYB review — resolve the entity, then pull registry details, officers, shareholders, group structure and financials into one dossier; company-snapshot gives a fast one-screen profile instead of the full dossier; prospect-list builds a target account list from firmographic filters, optionally enriched with contacts.

Where it is listed

Verified 2026-09-01.

Also installable without a directory:

Not listed: PulseMCP (their submissions are paused), mcp.so, and the VS Code / GitHub MCP gallery — that last one is a curated catalogue of a few hundred servers and is not an ingest of the Official MCP Registry, so publishing there does not reach it. Also not pursued: Perplexity's curated connector catalog (no self-serve submission exists — only the custom-connector route above) and Microsoft's MCP server certification for Copilot Studio (a real Partner Center pipeline, unlike Perplexity's — see PUBLISHING.md).

Maintainer notes — publishing, DNS verification and the remaining directories: PUBLISHING.md.

Links

• Homepage
• Smithery listing
• Privacy policy
• Terms

About this repository

Distribution metadata only (server.json, .mcp.json, plugin.json, .claude-plugin/, .cursor-plugin/, gemini-extension.json) so MCP directories and clients can discover the hosted server, plus the client-side extras that ship with it — the three skills and the Copilot Studio instruction block. MIT covers this metadata; the server implementation and the Global Database API are not open source.