Skip to main content
The context entry points covered so far all encode knowledge ahead of time — your schema, your rules, your documents, your examples. But some context lives in other systems and changes constantly: a CRM’s account state, a ticketing queue, a status page, an internal service. The Model Context Protocol (MCP) is an open standard for connecting exactly those systems to an AI agent, and your Upsolve agent can call external MCP servers at chat time to pull that live context into its answers.
MCP Connectors tab in the agent configuration showing connected external sources

​
Where MCP fits among the entry points

Each context entry point answers a different question: MCP is the entry point for anything the agent should fetch live rather than be told in advance. A question like “are there any open incidents affecting this customer?” can’t be answered from your data model — but an MCP connector to your incident tracker can.

​
Managing connectors

MCP connectors are configured at the project level (Project Settings → MCP Connectors) and are enabled for every agent by default. In an agent’s own MCP Connectors tab you can turn specific connectors off for that agent version — useful when a connector’s tools would be noise for a narrowly scoped agent. Connector changes are draft-scoped: they take effect once the agent version is saved.

​
The other direction: exposing Upsolve over MCP

This page is about MCP as a way to bring external context into your agent. Upsolve also uses MCP in the opposite direction — exposing your agent to MCP clients. Those are deployment topics, covered in their own guides:

Claude MCP Server

Explore your own data in Claude, with charts rendered in the conversation.

Agent MCP for Your Users

Expose your agent to end users or your own agent, scoped per user by token.

​
Next steps

Knowledge Base

Store static reference documents the agent searches on demand.

Skills

Package procedures the agent follows once live context arrives.