MCP server

The Sciple MCP server is a standalone Python service that exposes runbook authoring as tools an AI assistant can call. It speaks the Model Context Protocol over stdio and runs locally through Claude Code or Claude Desktop, so you can build and edit runbooks by talking to Claude.

Configuration

The server is configured through environment variables:

• SCIPLE_API_URL: the URL of your Sciple API.
• SCIPLE_API_TOKEN: a scoped personal access token.
• SCIPLE_TENANT_ID: the tenant the server acts within.

Tools Claude can call

The server exposes nine tools, grouped by what they do.

• Reading and creating runbooks: list_runbooks, get_runbook, create_runbook.
• Working with cells: add_cell, update_cell, delete_cell, reorder_cells.
• Lifecycle: promote_runbook, deprecate_runbook.

Safety model

Claude drafts and edits runbooks in conversation, but nothing executes or becomes a team standard until an engineer reviews and promotes it in the Sciple UI. The token Claude uses is scoped, not your full credentials, so the blast radius is limited to runbook authoring.

Example

You might ask Claude: create a runbook to scale down an ECS service, with a markdown cell explaining when to use it, a shell cell that lists services, and a shell cell that sets desired count to zero, then mark it reviewed. Claude drafts the runbook and its cells, and an engineer reviews and promotes it in the UI when it is ready to become a standard.

Built on open standards

The server is built on the open Model Context Protocol and the official mcp SDK.

What is next

• Runbooks covers cell types, targets, the lifecycle, and running a runbook.