MiMo2API

OAuth to OpenAI-compatible API router for Xiaomi MiMo Code Desktop. Bridge OAuth credentials into a universal API gateway to unlock Free Preview MiMo X models (MiMo-X-Pro-Preview & MiMo-X-Flash-Preview) across Claude Code, Codex, Antigravity, OpenCode and more.

Overview • Features • Quick Start • Models • Clients • Architecture • Guides • Disclaimer • License

What is MiMo2API?
Key Features
Quick Start
- Method 1: Local Daemon
- Method 2: Cloudflare Workers (Serverless)
Model Catalog & Aliases
Client Integration
Architecture & Structure
Advanced Guides
- Multi-Account Pooling & 429 Failover
- Environment Variables
Development & Testing
Disclaimer & Legal Notice
License

What is MiMo2API?

In September 2026, Xiaomi introduced the MiMo-X model family (MiMo-X-Pro-Preview and MiMo-X-Flash-Preview), offering advanced reasoning, multi-agent coordination, and coding capabilities through invitation-based beta testing in the Xiaomi MiMo Desktop application.

However, access to these free preview quotas is locked behind Xiaomi's OAuth authentication inside the desktop client. Developers could not consume these models as standard OpenAI-compatible endpoints (sk-...) in external coding agents, harnesses, or IDEs.

MiMo2API bridges this gap:

It reads and manages Xiaomi OAuth tokens (access_token and refresh_token) from your local MiMo Desktop configuration (auth.json), proactively refreshing access tokens before expiration.
It serves a standard OpenAI-compatible endpoint (/v1/chat/completions, /v1/models, /v1/health) with full Server-Sent Events (SSE) streaming and reasoning_content delta preservation.
It supports both a zero-dependency Local CLI Daemon (localhost:20128) and a standalone Cloudflare Worker (worker/worker.js) with multi-account round-robin rotation and automatic 429 quota exhaustion fallback.

Disclaimer: This tool is provided strictly for personal educational and interoperability testing purposes. Extracting or reusing OAuth credentials outside official desktop clients carries the risk of account suspension or ban. See Disclaimer & Legal Notice.

Key Features

Unlock Free Preview Models: Access MiMo-X-Pro-Preview, MiMo-X-Flash-Preview, and mimo-v2.5 series models.
Dual Operational Modes:
- Local Daemon: Runs on localhost:20128, auto-detecting desktop credentials with zero configuration.
- Cloudflare Worker: Standalone 20 KB bundle ready for copy-paste into Cloudflare Workers or Vercel Edge.
Multi-Account Round-Robin: Pool multiple refresh tokens; requests distribute automatically across accounts.
Intelligent 429 Failover: Automatically detects daily rate limits (Daily free limit reached, Try again in...), parks the exhausted account in cooldown, and retries the request transparently with the next available account.
Streaming SSE with Reasoning Delta: Preserves thinking blocks in delta.reasoning_content for real-time rendering in coding assistants.
Safe & Resource-Bounded:
- Client disconnect cancellation via AbortSignal (prevents wasted quotas when cancelling prompts).
- 120-second upstream timeout guards against stalled connections.
- 10MB request payload limit protects against buffer exhaustion attacks.
- Atomic token persistence using secure 0o600 file permissions.

Quick Start

Method 1: Local Daemon

Step 1: Verify Prerequisites

Install and log into the Xiaomi MiMo Desktop application. This saves your OAuth session in auth.json.

Install globally via npm:

npm install -g mimo2api

Or run directly with npx without installation.

Step 2: Check Authentication

Test if your local credentials are detected:

mimo2api check

Or via npx:

npx mimo2api check

Or when running from a local clone:

npm run build

node dist/cli.js check

Example output:

[MiMo2API] Checking local MiMo Desktop authentication...
Found configuration: C:\Users\<user>\AppData\Local\mimocode\data\auth.json
Found Xiaomi OAuth credentials:
  - Access Token: eyJhbGci... (1024 chars)
  - Refresh Token: dGhpcy1p... (256 chars)
  - Expires: 2026-09-18T23:59:00.000Z
Local authentication check passed!

Step 3: Start the Router

Run with default settings (port 20128, open access):

npx mimo2api start

Or specify a custom port:

npx mimo2api start --port 20128

Or enable optional API key authorization:

npx mimo2api start --port 20128 --key sk-my-secret-key

When running from source:

npm start

Your local endpoint is available at http://localhost:20128/v1.

Method 2: Cloudflare Workers (Serverless)

Deploy a 24/7 serverless gateway without keeping your computer running.

Step 1: Copy the Code

Open worker/worker.js and copy the entire file contents.

Step 2: Create Worker in Cloudflare Dashboard

Log into the Cloudflare Dashboard.
Go to Workers & Pages -> Create application -> Create Worker.
Set name to mimo2api and click Deploy.
Click Edit code, select all existing code, delete it, and paste the copied contents of worker/worker.js.
Click Deploy in the top right.

Step 3: Add Variables & Secrets

In your Worker, go to Settings -> Variables and Secrets and add:

Variable	Type	Required	Description
`MIMO_REFRESH_TOKEN`	Secret	Yes	Your MiMo refresh token. Multi-account supported: one token per line.
`API_KEY`	Secret	Optional	Client bearer key (e.g. `sk-mimo-key`).
`UPSTREAM_BASE_URL`	Text	Optional	Upstream URL (defaults to `https://api.xiaomimimo.com/v1`).

Click Save and deploy.

Your Cloudflare Worker API URL:

https://mimo2api.<your-subdomain>.workers.dev/v1

Model Catalog & Aliases

MiMo2API automatically maps requested model aliases to canonical upstream models:

Request Model Alias	Canonical Upstream Model	Description
`mimo-x-pro`, `mimo-x`, `gpt-4o`	`MiMo-X-Pro-Preview`	Advanced reasoning model for complex architectural coding
`mimo-x-flash`, `mimo-flash`	`MiMo-X-Flash-Preview`	Low-latency preview model for rapid edits and tool calls
`mimo-v2.5-pro`, `mimo-v2.5`	`mimo-v2.5-pro`	Xiaomi MiMo V2.5 base production model
`mimo-v2.5-flash`	`mimo-v2.5-flash`	Xiaomi MiMo V2.5 high-throughput model

Client Integration

9Router

Configuration file path:

~/.9router/db.json

Or configure via Web Dashboard under Providers -> Add Custom Provider:

Provider Type:

openai

Base URL:

http://localhost:20128/v1

API Key:

sk-mimo

Models:

mimo-x-pro, mimo-x-flash

Configuration entry for ~/.9router/db.json:

{
  "providers": [
    {
      "id": "mimo2api",
      "name": "MiMo2API",
      "type": "openai",
      "baseUrl": "http://localhost:20128/v1",
      "apiKey": "sk-mimo",
      "models": [
        "mimo-x-pro",
        "mimo-x-flash"
      ]
    }
  ]
}

Note: Both 9Router and MiMo2API default to port 20128. When running both locally on the same host, start MiMo2API on a different port (e.g. PORT=20129 npx mimo2api start) and point 9Router to http://localhost:20129/v1, or point 9Router to your Cloudflare Worker URL.

Aider

Configuration file path:

.aider.conf.yml

Add configuration:

openai-api-base: http://localhost:20128/v1
openai-api-key: sk-mimo
model: openai/mimo-x-pro

Antigravity (AGY)

Configuration file path:

~/.gemini/antigravity/antigravity.json

Add configuration:

{
  "modelProviders": {
    "mimo2api": {
      "type": "openai",
      "baseUrl": "http://localhost:20128/v1",
      "apiKey": "sk-mimo",
      "defaultModel": "mimo-x-pro"
    }
  }
}

Cherry Studio

Configuration file path:

~/.cherry-studio/config.json

Or configure via UI in Settings -> Providers -> OpenAI:

Custom Server Address:

http://localhost:20128/v1

API Key:

sk-mimo

Models:

mimo-x-pro

mimo-x-flash

Claude Code

Configuration file path (Global):

~/.claude/settings.json

Configuration file path (Project-level):

.claude/settings.json

Add configuration:

{
  "env": {
    "OPENAI_BASE_URL": "http://localhost:20128/v1",
    "OPENAI_API_KEY": "sk-mimo",
    "ANTHROPIC_MODEL": "mimo-x-pro"
  }
}

Then run:

claude

Cline

Configuration file path:

.vscode/settings.json

Add configuration:

{
  "cline.apiProvider": "openai-compatible",
  "cline.openAiBaseUrl": "http://localhost:20128/v1",
  "cline.openAiApiKey": "sk-mimo",
  "cline.openAiModelId": "mimo-x-pro"
}

Codex

Configuration file path:

~/.codex/config.toml

Add configuration:

[model]
provider = "openai"
base_url = "http://localhost:20128/v1"
api_key = "sk-mimo"
model_name = "mimo-x-pro"

Continue.dev

Configuration file path:

~/.continue/config.json

Add configuration:

{
  "models": [
    {
      "title": "MiMo-X Pro Preview",
      "provider": "openai",
      "model": "mimo-x-pro",
      "apiBase": "http://localhost:20128/v1",
      "apiKey": "sk-mimo"
    }
  ]
}

Cursor

Configuration file path (Global):

~/.cursor/User/settings.json

Configuration file path (Project-level):

.vscode/settings.json

Add configuration:

{
  "cursor.openaiBaseUrl": "http://localhost:20128/v1",
  "cursor.openaiApiKey": "sk-mimo",
  "cursor.model": "mimo-x-pro"
}

Or configure in Cursor Settings -> Models:

Toggle Override OpenAI Base URL: http://localhost:20128/v1
Set OpenAI API Key: sk-mimo
Add model: mimo-x-pro

DeepSeek Harness

Configuration file path:

agent.yaml

Add configuration:

llm:
  api_type: openai
  base_url: "http://localhost:20128/v1"
  api_key: "sk-mimo"
  model: "mimo-x-pro"
  temperature: 0.7

Hermes

Configuration file path:

~/.hermes/config.json

Add configuration:

{
  "llm": {
    "provider": "openai",
    "baseUrl": "http://localhost:20128/v1",
    "apiKey": "sk-mimo",
    "model": "mimo-x-pro"
  }
}

LibreChat

Configuration file path:

librechat.yaml

Add configuration:

endpoints:
  custom:
    - name: "MiMo2API"
      apiKey: "sk-mimo"
      baseURL: "http://localhost:20128/v1"
      models:
        default: ["mimo-x-pro", "mimo-x-flash"]
      titleConvo: true
      modelDisplayLabel: "MiMo"

MiMo Code CLI

Configuration file path (Linux / macOS):

~/.local/share/mimocode/mimocode.jsonc

Configuration file path (Windows):

%LOCALAPPDATA%\mimocode\data\mimocode.jsonc

Add configuration:

{
  "providers": {
    "mimo2api": {
      "type": "openai-compatible",
      "baseUrl": "http://localhost:20128/v1",
      "apiKey": "sk-mimo",
      "models": [
        "mimo-x-pro",
        "mimo-x-flash"
      ]
    }
  },
  "default_model": "mimo2api/mimo-x-pro"
}

NextChat (ChatGPT-Next-Web)

Configuration file path:

.env.local

Add configuration:

BASE_URL=http://localhost:20128
OPENAI_API_KEY=sk-mimo
CUSTOM_MODELS=-all,+mimo-x-pro,+mimo-x-flash

OmniRoute

Configuration file path:

~/.omniroute/providers.json

Add configuration via CLI:

omniroute provider add --id mimo2api --type openai --base-url http://localhost:20128/v1 --api-key sk-mimo --models mimo-x-pro,mimo-x-flash

Or add configuration to ~/.omniroute/providers.json:

{
  "providers": [
    {
      "id": "mimo2api",
      "name": "MiMo Preview Gateway",
      "type": "openai-compatible",
      "baseUrl": "http://localhost:20128/v1",
      "apiKey": "sk-mimo",
      "models": [
        "mimo-x-pro",
        "mimo-x-flash"
      ]
    }
  ]
}

OpenAI Compatible (Generic / SDKs)

Configuration file path:

.env

Add environment configuration:

OPENAI_BASE_URL="http://localhost:20128/v1"
OPENAI_API_KEY="sk-mimo"

Python SDK example:

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:20128/v1",
    api_key="sk-mimo"
)

response = client.chat.completions.create(
    model="mimo-x-pro",
    messages=[{"role": "user", "content": "Write quicksort in Python."}],
    stream=True
)

for chunk in response:
    content = chunk.choices[0].delta.content or ""
    print(content, end="", flush=True)

Node.js SDK example:

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "http://localhost:20128/v1",
  apiKey: "sk-mimo",
});

const response = await client.chat.completions.create({
  model: "mimo-x-pro",
  messages: [{ role: "user", content: "Write quicksort in TypeScript." }],
});

console.log(response.choices[0].message.content);

cURL command:

curl -X POST http://localhost:20128/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-mimo" \
  -d '{"model":"mimo-x-pro","messages":[{"role":"user","content":"Hello!"}]}'

OpenClaw

Configuration file path:

openclaw.json

Add configuration:

{
  "providers": {
    "mimo2api": {
      "type": "openai-compatible",
      "baseURL": "http://localhost:20128/v1",
      "apiKey": "sk-mimo",
      "models": [
        "mimo-x-pro",
        "mimo-x-flash",
        "mimo-v2.5-pro"
      ]
    }
  }
}

OpenCode

Configuration file path (Linux / macOS):

~/.local/share/opencode/opencode.jsonc

Configuration file path (Project-level):

opencode.jsonc

Add configuration:

{
  "providers": {
    "mimo2api": {
      "type": "openai-compatible",
      "baseUrl": "http://localhost:20128/v1",
      "apiKey": "sk-mimo",
      "models": [
        "mimo-x-pro",
        "mimo-x-flash"
      ]
    }
  },
  "default_model": "mimo2api/mimo-x-pro"
}

OpenHands (OpenDevin)

Configuration file path:

config.toml

Add configuration:

[llm]
model = "openai/mimo-x-pro"
base_url = "http://localhost:20128/v1"
api_key = "sk-mimo"

Roo Code

Configuration file path:

.vscode/settings.json

Add configuration:

{
  "roo-cline.apiProvider": "openai-compatible",
  "roo-cline.openAiBaseUrl": "http://localhost:20128/v1",
  "roo-cline.openAiApiKey": "sk-mimo",
  "roo-cline.openAiModelId": "mimo-x-pro"
}

Trae (ByteDance Agentic IDE)

Configuration file path:

~/.trae/config.json

Add configuration:

{
  "modelProviders": [
    {
      "name": "MiMo2API",
      "apiType": "openai",
      "endpoint": "http://localhost:20128/v1",
      "apiKey": "sk-mimo",
      "models": [
        "mimo-x-pro",
        "mimo-x-flash"
      ]
    }
  ]
}

Windsurf

Configuration file path:

~/.codeium/windsurf/model_config.json

Add configuration:

{
  "customOpenAI": {
    "endpoint": "http://localhost:20128/v1",
    "apiKey": "sk-mimo",
    "model": "mimo-x-pro"
  }
}

Architecture & Structure

MiMo2API/
├── src/
│   ├── types/
│   │   ├── auth.ts              # OAuth credentials, pool states, refresh contracts
│   │   ├── openai.ts            # OpenAI request/response/SSE schemas & chunk deltas
│   │   └── config.ts            # Server and Cloudflare Worker runtime configuration
│   ├── auth/
│   │   ├── token-store.ts       # Discovers and parses local auth.json with atomic writes
│   │   ├── oauth-client.ts      # Xiaomi OAuth 2.0 exchange & proactive TTL refresh
│   │   └── pool-manager.ts      # Multi-account round-robin pool with 429 cooldown backoff
│   ├── proxy/
│   │   ├── model-catalog.ts     # Model list & alias normalizer (mimo-x-pro -> MiMo-X-Pro-Preview)
│   │   ├── stream-transformer.ts# Real-time SSE transformer with reasoning_content support
│   │   └── completion-handler.ts# Core completion proxy with cancellation & failover retry
│   ├── server/
│   │   ├── routes.ts            # Route dispatcher (/v1/models, /v1/chat/completions, /v1/health)
│   │   └── http-server.ts       # Native Node.js 22 HTTP server with graceful shutdown
│   ├── cli/
│   │   └── index.ts             # CLI entrypoint (start, check)
│   └── worker/
│       └── index.ts             # Cloudflare Workers / Vercel Edge fetch entrypoint
├── worker/
│   └── worker.js                # Standalone zero-dependency Cloudflare Worker bundle (20 KB)
├── dist/
│   └── cli.js                   # Standalone Node.js CLI executable bundle (29 KB)
├── tests/                       # 36 automated unit & integration tests
├── scripts/
│   └── build.ts                 # esbuild build pipeline
├── .github/
│   └── workflows/
│       └── test.yml             # Automated CI workflow with skip test flags
├── package.json
└── tsconfig.json

Advanced Guides

Multi-Account Pooling & 429 Failover

If an account reaches its preview quota limit, MiMo2API automatically shifts traffic to the next account without interrupting your active session.

In Cloudflare Workers or your local environment, pass multiple refresh tokens separated by newlines in MIMO_REFRESH_TOKEN:

MIMO_REFRESH_TOKEN="token_account_1
token_account_2
token_account_3"

Workflow:

Round-Robin: Requests cycle across active accounts to distribute load.
Quota Detection: When an account encounters HTTP 429 (Daily free limit reached or Try again in 2h 30m), the router parses the duration and parks that account in cooldown.
Transparent Retry: The current request is immediately retried using the next available account in the pool.
Health Diagnostics: Query /v1/health to monitor active versus cooldown accounts.

Environment Variables

Variable	Mode	Default	Description
`PORT`	Local	`20128`	Local HTTP daemon port
`HOST`	Local	`127.0.0.1`	Local bind address
`API_KEY`	Both	(None)	Client authorization bearer token
`MIMO_REFRESH_TOKEN`	Both	(From auth.json)	Newline-separated list of refresh tokens
`MIMOCODE_HOME`	Local	(Auto)	Custom directory to look for `auth.json`
`UPSTREAM_BASE_URL`	Both	`https://api.xiaomimimo.com/v1`	Upstream MiMo inference endpoint
`OAUTH_TOKEN_URL`	Both	`https://account.xiaomi.com/oauth2/token`	Upstream Xiaomi OAuth token endpoint

Development & Testing

Run typecheck:

npm run typecheck

Run automated test suite:

npm test

Build standalone bundles:

npm run build

Continuous Integration (CI)

Every commit and pull request runs automated verification through the Test & Verify GitHub Actions workflow (Node.js 22):

Typecheck: npm run typecheck
Unit & Integration Tests: npm test
Bundle Verification: npm run build

Skipping Automated Tests

To skip the test suite on specific commits (e.g. for docs updates, asset additions, or fast iterations while still verifying types and builds), include any of the following tags in the commit message:

[skip test] or [skip tests]
[skip-test] or [skip-tests]

When triggered manually via Actions -> Run workflow, check the Skip running automated tests input option.

Disclaimer & Legal Notice

IMPORTANT: Please read this notice carefully before using or deploying MiMo2API.

Educational & Research Purposes Only: This project is developed and distributed exclusively for personal educational, research, and non-commercial API interoperability testing purposes.
Risk of Account Suspension or Ban: Extracting OAuth tokens, managing session credentials, or routing automated API requests outside the official Xiaomi MiMo Desktop client may violate Xiaomi's Terms of Service and End User License Agreements. Xiaomi may rate-limit, revoke tokens for, suspend, or permanently ban your Xiaomi account without prior warning.
No Warranty & No Guarantee: The author and contributors make no claims, promises, or guarantees regarding the safety, status, or longevity of your Xiaomi account. This software is provided "AS IS", without warranty of any kind, express or implied.
Assumption of Risk: You assume full and sole responsibility for any outcomes or damages resulting from using this software, including but not limited to account termination, loss of API access, or data loss. Use strictly at your own risk.
Trademark Attribution: All product names, logos, and brands (such as "Xiaomi", "MiMo", "MiMo-Code") are trademarks or registered trademarks of their respective owners (Xiaomi Inc.). MiMo2API is an independent open-source project and is neither affiliated with, maintained by, nor endorsed by Xiaomi Inc.

License

This project is licensed under the GNU General Public License v3.0 (GPLv3). See the LICENSE file for details.

Name		Name	Last commit message	Last commit date
.github/workflows		.github/workflows
docs/specs		docs/specs
public		public
scripts		scripts
src		src
tests		tests
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
package-lock.json		package-lock.json
package.json		package.json
tsconfig.json		tsconfig.json

Navigation Menu

Repository files navigation

MiMo2API

Table of Contents

What is MiMo2API?

Key Features

Quick Start

Method 1: Local Daemon

Step 1: Verify Prerequisites

Step 2: Check Authentication

Step 3: Start the Router

Method 2: Cloudflare Workers (Serverless)

Step 1: Copy the Code

Step 2: Create Worker in Cloudflare Dashboard

Step 3: Add Variables & Secrets

Model Catalog & Aliases

Client Integration

Architecture & Structure

Advanced Guides

Multi-Account Pooling & 429 Failover

Environment Variables

Development & Testing

Continuous Integration (CI)

Skipping Automated Tests

Disclaimer & Legal Notice

License

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

MiMo2API

Table of Contents

What is MiMo2API?

Key Features

Quick Start

Method 1: Local Daemon

Step 1: Verify Prerequisites

Step 2: Check Authentication

Step 3: Start the Router

Method 2: Cloudflare Workers (Serverless)

Step 1: Copy the Code

Step 2: Create Worker in Cloudflare Dashboard

Step 3: Add Variables & Secrets

Model Catalog & Aliases

Client Integration

Architecture & Structure

Advanced Guides

Multi-Account Pooling & 429 Failover

Environment Variables

Development & Testing

Continuous Integration (CI)

Skipping Automated Tests

Disclaimer & Legal Notice

License

About

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages