Set up Apideck MCP Server

The Apideck MCP Server connects an AI client, through seven unified APIs, to more than 200 SaaS applications across accounting, CRM, HRIS, ATS, file storage, ecommerce, and issue tracking. Because several of these categories allow write and destructive operations, a deliberate setup pays off before connecting production third-party accounts.

Create credentials in the Apideck dashboard

An Apideck account provides an API key and an app ID; both are required as headers on every call to the MCP server. Teams serving multiple end customers through their own application can additionally set a consumer ID per call to attribute requests to a specific customer.

Choose between the hosted endpoint and a local server

For a quick start, it's enough to configure the hosted endpoint https://mcp.apideck.dev/mcp directly in the MCP client. To run the server yourself instead, start it locally with npx -y @apideck/mcp start; a prebuilt bundle for Claude Desktop is also available in the repository's GitHub releases.

Pick the right operating mode

For most agents, the default "Dynamic" mode fits best, since it initially loads only a few meta-tools and pulls in more only when actually needed. Only agents working very narrowly and exclusively with the Apideck API benefit from "Static" mode with all roughly 360 tools loaded upfront; for lean, programmatic control, "Code" mode is the better fit instead.

Authorize third-party integrations and scope access narrowly

Each connected third-party application is authorized once via OAuth through Apideck Vault; Apideck then handles token storage and refresh. Set the available scope filter as narrowly as possible, for example read-only access for exploratory tasks, and enable write or destructive operations only once a concrete, reviewed use case requires it. The API key and app ID belong only in protected environment variables or the MCP client's secure configuration, never in prompts or repositories.