MCP Server

Tediware now has an official Model Context Protocol (MCP) server at https://tediware.com/mcp, so a coding agent like Claude Code or Cursor can read your configuration, trace EDI transactions as they run through the platform, and send and receive data using an API key.

Impact: nothing changes unless you connect an agent. To try it, point your coding agent at the MCP documentation and give it an API key from the API Keys page. The launch post has a demo video.

• Standard and read-only keys both work. A sandbox key is rejected with 403.
• A read-only key does not see the tools that change anything: partner_send, partner_receive, transaction_resend, transaction_redeliver and suggestion_submit. It still reads your transactions in full, including their business data.
• The write tools are marked readOnlyHint: false, so your agent framework can ask you to confirm before it runs one. Create one key per agent and use a read-only key unless the agent’s job includes sending.
• MCP calls count against the same rate limits as the equivalent API endpoints.
• Most agents launch MCP servers over stdio. tedi mcp serve, from the tedi CLI, bridges them to the hosted server and keeps the key out of the agent’s configuration file: claude mcp add tediware -- tedi mcp serve.
• The configuration tools (partner_*, connection_*, mapping_* and the rest) are in preview, and their output shapes may change.

Setup, the full tool list and error codes are in the MCP documentation.

This update was written with AI assistance. It has been reviewed and edited by a human.