AI Hub MCP Server

AI Hub MCP Server

Use this skill for the full IRIS-to-MCP path: toolsets in IRIS, %AI.MCP.Service, iris-mcp-server, transport selection, discovery, and security.

When to Use

• Publishing IRIS tools to Claude Desktop through stdio
• Exposing remote MCP endpoints over HTTP or HTTPS
• Figuring out iris-mcp-server configuration, discovery, and namespacing
• Troubleshooting tool visibility, authentication failures, or TLS setup

Core Workflow

1. Build ObjectScript tools or toolsets that represent the IRIS capabilities.
2. Expose them through a %AI.MCP.Service subclass.
3. Create the corresponding MCP application in IRIS.
4. Configure iris-mcp-server with the right transport and endpoint mapping.
5. Set up both transport-to-IRIS credentials and endpoint-user credentials.
6. Verify discovery, tool namespacing, and security behavior.

Critical Rules

• Treat iris-mcp-server authentication as two separate layers.
• Do not assume stdio and HTTP/HTTPS support the same security options.
• Expect service IDs to prefix discovered tool names.
• Pair this skill with ai-hub-toolsets for IRIS-side tool design and ai-hub-policies for authorization or audit behavior.

Common Mistakes

• Reusing one credential set for both gateway access and endpoint user identity
• Forgetting that OAuth passthrough is only relevant to HTTP/HTTPS transport
• Missing tool-name prefixes caused by service namespacing
• Debugging discovery issues without checking iris_status, logs, or endpoint config first

References

• MCP server reference

Source: intersystems-community/ai-hub-dev-template — distributed by TomeVault.