modify-agent

To add MCP tools to a LangGraph agent, you integrate Databricks-hosted or external MCP servers via the LangChain Databricks MCP client, wiring them using DatabricksMCPServer or DatabricksMultiServerMCPClient and steering edits to agent_server/agent.py.

The correct way to update system instructions in LangGraph is by prepending messages to the conversation history, rather than passing unsupported parameters to the create_agent function, ensuring safe instruction customization without breaking SDK patterns.

Yes, you can change model selection by steering edits to agent_server/agent.py to switch to an appropriate model endpoint, ensuring the modification satisfies functional requirements for serving and evaluation without breaking tracing or integration expectations.

Yes, this approach supports integrating Databricks SDK usage by applying edits to agent_server/agent.py and wiring tools through DatabricksMCPServer and DatabricksMultiServerMCPClient, maintaining correct SDK patterns during agent modification.

You should prepend messages because using unsupported create_agent parameters for system instructions breaks SDK patterns, whereas the message-prepending pattern applies system instructions correctly in LangGraph workflows without disrupting tracing or integration expectations.

The best way to modify agent code safely is to steer edits to agent_server/agent.py, updating core logic, model selection, and tool integrations while using the correct message-based system instruction pattern to avoid breaking SDK patterns.