問題文
A knowledge source must be available to several agents built by different teams, without each team writing its own client code for it. Which approach fits?
選択肢
- Copy the source into each agent's system instructions
- Expose the source through an MCP server so any MCP-capable agent can discover and call it through the same protocol
- Fine-tune a separate model for each team on the contents of the source
- Give each team the connection string for the underlying database and let them query it directly, since a single shared credential keeps every team reading one consistent view of the source