I am attempting to connect to my custom MCP server to the new new Agent Builder, but when I supply my MCP server URL and authorization token, I get an error saying “Unable to load tools for this server”
My browser console shows error: api.openai.com/v1/dashboard/responses/mcp/list_actions:1 Failed to load resource: the server responded with a status of 424 ()
{
“error”: {
“message”: “Error retrieving tool list from MCP server: ‘http://{{server_url_here}}:4200/mcp’. Http status code: 424 (Failed Dependency)”,
“type”: “external_connector_error”,
“param”: “tools”,
“code”: “http_error”
}
}
I have confirmed I am able to successfully inspect my MCP server via the standard MCP inspector tool
I am encountering the same issue; I tried a few remote tools from smithery, including just the ‘weather’ GPT5 advised me to adjust the url’s to SSE versions, no avail.
Looking forward to these bugs being fixed so the workflows are usable!
Same issue here, mcp inspector is able to discover tools and run tools , but AgentBuilder is not able to discover the tool on my same remote host mcp server
Was working till yesterday, OpenAI anyone looking into this ? I am hoping to do a important demo on this tonight and now the flow is broken
I recently tested it and found that it works on MCP servers, both with authentication enabled (using bearer tokens) and without any authentication. Please let me know the details if you’re having a different experience!
Are you using an MCP server that supports SSE or streamable HTTP ? You can’t specify the option in the modal so I figured that it would automatically detect which protocol is being used … but maybe that’s the issue ?
I’m experiencing the same 424 “Failed Dependency” error with OpenAI Agent Builder when trying to connect my MCP server. After extensive testing, I confirmed my MCP server works perfectly with direct API calls (initialize and tools/list both return valid responses), but Agent Builder’s internal processing fails with 424 errors on list_actions requests. The issue is in OpenAI’s internal validation layer (https://api.openai.com/v1/dashboard/responses/mcp/list_actions), not in my MCP server configuration or authentication. This appears to be a OpenAI Agent Builder bug affecting multiple users with working MCP servers. Hopefully they will fix it soon.