Goal: let Claude, ChatGPT or an AI agent you already run use your system safely, through an MCP (Model Context Protocol) server, which calls no AI model itself.

For a Swiss industrial client I built the MCP server and AI assistant of their data platform from the start, inside their engineering team.

Starter: one system via its API, up to 3 read-only tools, one service account or API key, setup guide.

Standard: up to 6 tools incl. 2 writes (preview, then confirm), per-user permissions, audit log, API or database access, one deployment.

Advanced: 2 systems, up to 10 tools incl. 4 writes, SSO, rate limits, searchable audit log, test cases, docs.

Not included: the AI tool, agent or chat; you bring your MCP client, such as Claude or Cursor. A chat assistant in your app is my AI Assistant service.

Done when:

each tool returns the agreed results in your MCP client on test data;

from Standard, writes wait for their confirm, a test user is refused what they may not do, and calls are logged;

you have the source and setup guide.

NDA on request.

Packages: Starter $600, Standard $1,400, Advanced $2,800.