muapi-chat-agents

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Muapi chat agents

Muapi 聊天 Agent

Muapi serves coding agents natively, so no translation proxy is needed:

Agent	Protocol	Base URL	Models endpoint
Claude Code	Anthropic Messages	`https://api.muapi.ai/anthropic`	`GET /anthropic/v1/models`
Codex CLI	OpenAI Responses	`https://api.muapi.ai/openai/v1`	`GET /openai/v1/models`

Usage bills against the user's Muapi credits at the same per-model prices as the Muapi API. Auth is always the user's Muapi API key, read from

MUAPI_API_KEY

(never paste or commit it).

Muapi 原生支持编程 Agent，因此无需协议转换代理：

Agent	协议	Base URL	模型端点
Claude Code	Anthropic Messages	`https://api.muapi.ai/anthropic`	`GET /anthropic/v1/models`
Codex CLI	OpenAI Responses	`https://api.muapi.ai/openai/v1`	`GET /openai/v1/models`

使用量将按与 Muapi API 中各模型相同的价格，计入用户的 Muapi 额度。认证始终使用用户的 Muapi API key，该密钥从

MUAPI_API_KEY

读取（切勿粘贴或提交它）。

Rules

规则

Do not put the full request URL in the base URL. Agents append
```
/v1/messages
```
or
```
/responses
```
themselves.
Always read model ids from the models endpoint. Never rely on a remembered list.
Put provider settings in user-level config, never in project files.

Authenticate with

Authorization: Bearer <key>

(Claude Code's

ANTHROPIC_AUTH_TOKEN

does this) or

x-api-key

A request is checked against the wallet before it runs, so the balance must cover a worst-case request (a large prompt plus the full output limit on a big model can hold a dollar or more; the unused part is returned).

不要在 Base URL 中放入完整的请求 URL。Agent 会自行追加
```
/v1/messages
```
或
```
/responses
```
。
始终从模型端点读取模型 ID，绝不要依赖记忆中的列表。
将 provider 设置放在用户级配置中，绝不要放在项目文件中。
使用
```
Authorization: Bearer <key>
```
（Claude Code 的
```
ANTHROPIC_AUTH_TOKEN
```
采用此方式）或
```
x-api-key
```
进行认证。
请求在执行前会检查钱包余额，因此余额必须足以覆盖最坏情况的请求成本（大型 prompt 加上大型模型的完整输出上限，可能占用 1 美元或更多；未使用的部分会退还）。

Setup

设置

export MUAPI_API_KEY=your_key

(PowerShell:

$env:MUAPI_API_KEY = "your_key"

and

setx MUAPI_API_KEY "your_key"

List models:

bash

curl -s -H "Authorization: Bearer $MUAPI_API_KEY" https://api.muapi.ai/anthropic/v1/models
curl -s -H "Authorization: Bearer $MUAPI_API_KEY" https://api.muapi.ai/openai/v1/models

(Windows PowerShell: use

curl.exe

export MUAPI_API_KEY=your_key

（PowerShell：

$env:MUAPI_API_KEY = "your_key"

和

setx MUAPI_API_KEY "your_key"

）。

列出模型：

bash

curl -s -H "Authorization: Bearer $MUAPI_API_KEY" https://api.muapi.ai/anthropic/v1/models
curl -s -H "Authorization: Bearer $MUAPI_API_KEY" https://api.muapi.ai/openai/v1/models

（Windows PowerShell：使用

curl.exe

。）

Claude Code

Add to

~/.claude/settings.json

(or

.claude/settings.local.json

, never the committed project file):

json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.muapi.ai/anthropic",
    "ANTHROPIC_AUTH_TOKEN": "<your Muapi API key>",
    "ANTHROPIC_MODEL": "claude-sonnet-5-5"
  }
}

Or export

ANTHROPIC_BASE_URL

and

ANTHROPIC_AUTH_TOKEN="$MUAPI_API_KEY"

in the shell before starting

claude

settings.json

env values are literal, so prefer the shell exports to keep the key out of files. Run

/status

inside Claude Code to confirm the base URL, and

claude --debug

to see requests.

将以下内容添加到

~/.claude/settings.json

（或

.claude/settings.local.json

，切勿放入已提交的项目文件）：

json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.muapi.ai/anthropic",
    "ANTHROPIC_AUTH_TOKEN": "<your Muapi API key>",
    "ANTHROPIC_MODEL": "claude-sonnet-5-5"
  }
}

或者在启动

claude

前，在 shell 中导出

ANTHROPIC_BASE_URL

和

ANTHROPIC_AUTH_TOKEN="$MUAPI_API_KEY"

。

settings.json

中的 env 值为字面值，因此优先使用 shell 导出变量，以免将密钥保存在文件中。在 Claude Code 中运行

/status

以确认 Base URL，运行

claude --debug

以查看请求。

Uncensored (abliterated) models in Claude Code

Claude Code 中的无审查（abliterated）模型

GET /anthropic/v1/models

also lists the tool-capable abliterated models. Use the same base URL and key, pick one with

claude --model <id>

, and set

ANTHROPIC_DEFAULT_HAIKU_MODEL

and

ANTHROPIC_SMALL_FAST_MODEL

to that same id (otherwise Claude Code's background requests 404). Warn the user: quality varies by model (smaller models can skip steps on multi-file tasks), there is no prompt caching so every turn re-bills the whole conversation, and output is capped at 16,384 tokens per response. Codex CLI can use them too through

/openai/v1

(set

model

to the id); this path is tested with real Codex CLI sessions (file write, shell and

apply_patch

tool calls on qwen, mimo and glm), but quality varies by model.

GET /anthropic/v1/models

也会列出支持工具调用的无审查（abliterated）模型。使用相同的 Base URL 和密钥，通过

claude --model <id>

选择一个模型，并将

ANTHROPIC_DEFAULT_HAIKU_MODEL

和

ANTHROPIC_SMALL_FAST_MODEL

设置为同一个 ID（否则 Claude Code 的后台请求会返回 404）。提醒用户：模型质量存在差异（较小的模型在多文件任务中可能会跳过步骤），不支持 prompt 缓存，因此每轮都会重新计费整段对话，并且每次响应的输出上限为 16,384 tokens。Codex CLI 也可以通过

/openai/v1

使用这些模型（将

model

设置为对应 ID）；此路径已使用真实的 Codex CLI 会话进行测试（qwen、mimo 和 glm 上的文件写入、shell 与

apply_patch

工具调用），但质量因模型而异。

Codex CLI

~/.codex/config.toml

toml

model = "gpt-5-6-sol"
model_provider = "muapi"

[model_providers.muapi]
name = "Muapi"
base_url = "https://api.muapi.ai/openai/v1"
env_key = "MUAPI_API_KEY"
wire_api = "responses"

env_key

names the variable that holds the key; it is not the key. Set

model

explicitly to an id from the models endpoint. Provider ids

openai

ollama

and

lmstudio

are reserved, so use

muapi

. Run

codex doctor

to verify. Optional:

model_reasoning_effort = "high"

~/.codex/config.toml

：

toml

model = "gpt-5-6-sol"
model_provider = "muapi"

[model_providers.muapi]
name = "Muapi"
base_url = "https://api.muapi.ai/openai/v1"
env_key = "MUAPI_API_KEY"
wire_api = "responses"

env_key

指定存放密钥的环境变量名称，而不是密钥本身。请将

model

明确设置为模型端点返回的 ID。Provider ID

openai

、

ollama

和

lmstudio

已被保留，因此请使用

muapi

。运行

codex doctor

进行验证。可选设置：

model_reasoning_effort = "high"

。

Troubleshooting

故障排查

Problem	Fix
401	Header must be `Authorization: Bearer <key>` or `x-api-key` . Check the key at https://muapi.ai/dashboard
402 / insufficient credit	Balance cannot cover the worst case for that request; top up, or lower the output limit
404 model not found	Use an id from the models endpoint
`/v1/v1/messages` in errors	Remove the path suffix from the base URL
Claude Code asks to log in to Anthropic	Quit the terminal, export the variables, reopen it, start `claude` from that shell
Codex ignores the provider	Move settings to `~/.codex/config.toml` (user level)
5xx / provider error	Retry; the proxy already retries a backup route once

问题	解决方法
401	请求头必须是 `Authorization: Bearer <key>` 或 `x-api-key` 。请在 https://muapi.ai/dashboard 检查密钥
402 / 额度不足	余额无法覆盖该请求的最坏情况；请充值或降低输出上限
404，未找到模型	使用模型端点返回的 ID
错误中显示 `/v1/v1/messages`	从 Base URL 中移除路径后缀
Claude Code 要求登录 Anthropic	退出终端，导出环境变量，重新打开终端，并从该 shell 启动 `claude`
Codex 忽略 provider	将设置移至用户级的 `~/.codex/config.toml`
5xx / provider 错误	重试；代理已自动重试一次备用路由