← Help Center· General gateway

Access all Agents at once with magpie

magpieIt is a gateway that runs on your local machine and can convert between OpenAI, Anthropic, and Gemini protocols. Pick us up with it, just click a link—— Later Claude Code, Codex, Pi, OpenCode, Cursor Our models will appear in other tools.

Who is this path for? People who are already using multiple Agents and are willing to install a local gateway.
If you just want to use AI to chat without fussing with tools, just use ourCommand line clientThat’s fine - you don’t need to install a gateway.

1. One-click access (recommended)

Click the button below. magpie will pop up a confirmation window, **listing "where the prompt word and key will be sent"**, It won't be added until you click OK - nothing will be saved until you do.

Add to magpie

This linkKey not included, so it can be shared publicly. When adding magpie will let you paste the key yourself.

2. Add manually (equivalent approach)

If you don't want to use the link, manually add a custom supplier to the magpie:

Fill in whatValue
NameNewcclm
OpenAI compatible address https://sccclm.com/v1
Anthropic compatible addressLeave Empty(We do not provide Anthropic protocol)
KeyYoursk-*
Models newcclm、newcclm-fast、newcclm-coding、newcclm-free

Or use the command line:

magpie import 'magpie://import?name=Newcclm&chat=https://sccclm.com/v1&models=newcclm,newcclm-fast,newcclm-coding,newcclm-free'

3. What happens after you add

Our model will appear in each Agent's model selection list, writtennewcclm/newcclm-fast (supplier id/model nameformat next to it). You can:

The session level takes effect. magpie writes the configuration into the Agent's own configuration file (settings.json / config.tomletc. The session that is already running will not change——Open a new session to take effect.

4. Don't match it like this

Anthropic address field Do not enter our address

We only offer OpenAI compatible protocols. Bundlehttps://sccclm.com/v1If you fill in the Anthropic column, it will not connect. Leave EmptyThat’s it – magpie will automatically do the protocol conversion.

OpenAI address must have/v1

# Wrong
https://sccclm.com

# Correct
https://sccclm.com/v1

Do not enter the real model name

# Error - we only recognize our stall name
gpt-4o
glm-5.3-flash

# Correct
newcclm
newcclm-fast
newcclm-coding
newcclm-free

Do not place the import link with the key on a public page

The buttons on this page areKey not includedcan be made public. If you want to generate a keyed version, only put it on the page "Just issued a key to a user"—— The key is equal to the plaintext in the link.

5. Test it

First confirm that our interface is working separately, and then look at the configuration of magpie:

curl https://sccclm.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
-d '{"model":"newcclm-fast","messages":[{"role":"user","content":"Hello"}]}'

Agent reports 502/Can't connect, but there's nothing on magpie's routing page

This is a pitfall specifically mentioned in the magpie documentation: Agent software puts127.0.0.1:3425The request was taken and transferred elsewhere, the request did not reach magpie at all.

Just writing 127.0.0.1 in the "Bypass List" may not be enough (Agents such as Codex will still use proxies). Add two local direct connection rules to the proxy software and place them in front of other rules:

# Surge
IP-CIDR,127.0.0.0/8,DIRECT,no-resolve
IP-CIDR6,::1/128,DIRECT,no-resolve

# Clash / mihomo's rules: Next
- IP-CIDR,127.0.0.0/8,DIRECT,no-resolve
- IP-CIDR6,::1/128,DIRECT,no-resolve

6. How to undo

Select the model that comes with Agent in magpie's model selection, or run:

magpie claude default
magpie codex default

magpie will delete the content it writes, and replace your replaced value (such as the originalANTHROPIC_BASE_URL) to restore.

Still not working

1. Screenshot of magpie's "Routing" page (check whether the request has gone to the gateway)
2. The output of curl above
3. Which Agent do you use and what version of magpie

Back to Help Center · DeepSeek Harness Configuration →