IBM watsonx Orchestrate — Console Agent Setup

The connection, MCP toolkit and Python tool are all created by the CLI (make all). Only the agent itself is created in the console, because of the CLI limitation described in known-issues.md.

Field values below come from agents/neo4j_explorer.yaml.

Steps

  1. Open the Orchestrate console and go to Build → Agent Builder.

  2. Choose Create agent → Create from scratch. (Templates and LangGraph import are also offered here.)

  3. Set the fields:

Field Value

Name

neo4j_explorer

Description

Answers questions about companies, people, and investments using a Neo4j knowledge graph.

Model

bedrock/openai.gpt-oss-120b-1:0

Style

Default

  1. Paste the instructions block from the YAML file into Instructions.

  2. Open Toolset → Add tool + local instance and add:

    • the tools from the neo4j_local_mcp toolkit (get-schema, read-cypher)

    • the get_investments tool

  3. Save, then test in the preview chat (draft).

  4. Deploy to promote the agent to live. Live execution uses the live credentials of the neo4j_local_creds connection, so verify at least one query after deploying.

Adding the MCP server through the console instead

If you prefer not to use make mcp-toolkit, the same server can be registered from the UI: Toolset → Add tool + → Add from file or MCP server → Add local MCP server.

Field Value

Server name

neo4j_local_mcp

Description

Neo4j companies knowledge graph: schema inspection and read-only Cypher

Install command

uvx mcp-neo4j-cypher (package mcp-neo4j-cypher)

Connection

neo4j_local_creds