Backstop MCP - Technical Manual

1 min read

Overview

Technical reference for developers and architects working with Backstop MCP. For deployment and configuration, see the Operator Guide. For a short introduction, see the overview.

Backstop MCP is a pure MCP server: it exposes 44 tools that AI clients call on demand. It does not sync or index Backstop data. Apart from short-lived in-memory caches and the caller's own system-user record saved at login, every call reads or writes Backstop live, as the connected user.

Documentation

Document	What it covers
Architecture	Components, HTTP surface, request lifecycle, party resolution, the Backstop client, paging, caching, persistence, observability
Features	What the server can do, feature by feature: the questions each answers, its behaviour, and what is not supported
Tools	One line per tool, with the file that implements it and what it returns
Backstop Data Model	For developers changing the server: how the Backstop REST API behaves, and Backstop's standard records, fields and errors
Authentication	OAuth, the login form, token lifetimes, reconnecting, credential encryption
Security	Security checklist (section 7) for a self-managed install; Login Throttling and Encryption Key; what is logged in Architecture: Logs
Permissions	Every call runs as the connected user, and Backstop applies that user's own permissions; the server has no permission model of its own. See Authentication: How a Tool Call Authenticates and Features: Safeguards
FAQ	Common questions

Source

Path	Contents
`src/backstop_mcp/`	The server
`server/tools/registry.py`	The registered tools
`server/instructions.py`	The instructions every MCP client receives
`config.py`	Every setting
`packages/mcp-credential-auth`	Shared OAuth server, token storage, login throttling
`deploy/helm-charts/backstop-mcp`	Helm chart
`AGENT_README.md`	How features are structured, and the rules for adding one