MCP Servers
Model Context Protocol (MCP) is the open standard for connecting AI assistants like Claude to your tools, data, and systems. An MCP server exposes a set of resources or actions; Claude connects to it and can read those resources or invoke those actions inside conversations.
What RBA ships here
Two kinds of MCP servers live in this catalog:
- Integration servers — Connect Claude to client systems (a Sitecore instance, an internal API, a Postgres database). The server runs locally on a developer's machine or remotely in a private network.
- Workflow servers — Wrap an internal RBA process (a project-tracking tool, a deployment pipeline) so Claude can read status or trigger actions.
When to reach for MCP
MCP makes sense when the data or action lives outside the codebase and Claude needs live access to it during a conversation. Reach for an MCP server when:
- The data changes too often to bake into a skill (live database state, current ticket status, build pipeline runs).
- The action is one Claude should be able to perform during a conversation (kicking off a deploy, running a query, fetching a record).
- The integration is reusable across many projects — write the MCP server once, install it everywhere.
For one-shot lookups or static reference data, a skill or resource is usually a better fit than spinning up an MCP server.
Security note
MCP servers run with the privileges of the account that starts them. Review the configuration of any server before installing — especially one that connects to production systems. RBA's MCP servers ship with default-deny configurations; you opt in to specific scopes per project.