Connect Magpie to ModelGo: Setup & Troubleshooting

Magpie manages model selection for multiple Agents. Add ModelGo as a custom provider, then assign its models to your Agents.

1. Prepare your API key and model

Install Magpie and your Agent. Create an API key in the ModelGo console, and check your balance, budget, and model permissions.

Use a ModelGo logical model ID from the model catalog or query models available to your key:

bash

export MODELGO_API_KEY="mgk_your_API_key"
curl https://api.modelgo.ai/v1/models \
  -H "Authorization: Bearer $MODELGO_API_KEY"

Use a value from data[].id. This environment variable is for the terminal check; enter the key separately in Magpie.

2. Add ModelGo as a custom provider

Open Providers → Add provider → Custom.

Magpie provider sheet with the Custom entry at the bottom; Chinese UI example

*Official Chinese UI example. Choose Custom to add ModelGo.*

Setting	Value
Name	`ModelGo`
OpenAI-compatible URL	`https://api.modelgo.ai/v1`
Anthropic-compatible URL	`https://api.modelgo.ai`
API key	Your ModelGo API key

Save. Do not append /chat/completions or /messages. The selected model must support the Agent's API entry point; see model protocols and parameters.

3. Select a model for your Agent

Choose which models to expose under the ModelGo provider. Open Agents, then select a ModelGo model in the target Agent's model picker.

Magpie Agents page with a model picker for each Agent; Chinese UI example

*Official Chinese UI example showing picker locations; choose a model under your ModelGo provider.*

Start a new Agent session. For Codex, restart the application as well. Magpie writes the gateway and model settings into the Agent's configuration. Keep Magpie running while using its models; closing the window can leave it running in the menu bar or tray.

4. Verify

Send a short message in the new session. Check Magpie's Routing page to confirm that the request used your ModelGo model, or inspect the ModelGo call logs. Verify tools, files, and image understanding separately with small tasks.

Troubleshooting

Issue	Action
`401`	Check the key in Magpie for validity, missing characters, or whitespace.
`402`	Check balance and workspace, member, or key budgets.
`403`	Check the error reason, model permissions, and account status.
`404`	Check the base URL and avoid duplicate `/v1` or full request paths.
`model_protocol_not_offered`	Choose a model supporting the requested API entry point.
`429`	Follow `Retry-After` or retry with backoff.
Old model still used	Start a new session; restart Codex too.
No routing record	Keep Magpie running and check whether a proxy intercepts `127.0.0.1:3425`.

Keep X-Request-ID for troubleshooting. See errors and rate limits.