mock-ollama

Mock Ollama

mock-ollama Uniformly proxy the real LLM interface to the local machine.

It accepts the OpenAI Chat Completions, Anthropic Messages, and OpenAI Responses protocols;/api/version、/api/tags、/api/show It only provides Ollama discovery/metadata compatibility,does not support Ollama /api/chat and /api/generate。

Installation

npm install -g mock-ollama
mock-ollama -h

Cursor temporary public network mode additionally requires installing cloudflared。

Start the main service

When the upstream is OpenAI Chat Completions:

mock-ollama \
  --url "https://open.bigmodel.cn/api/paas/v4" \
  --apikey "你的上游密钥" \
  --api-style chat

Environment variables can also be used:

export MOCK_OLLAMA_BASE_URL="https://api.example.com/v1"
export MOCK_OLLAMA_API_KEY="你的上游密钥"
mock-ollama --api-style responses --bridge

By default listens on http://localhost:11434。

Protocols and bridge

Client protocol	Local route	Typical client
Chat Completions	/v1/chat/completions	Cursor, OpenAI-compatible SDK
Anthropic Messages	/v1/messages	Claude Code
OpenAI Responses	/v1/responses	Codex、GPT-5.6

Parameter	Description
--api-style auto|anthropic|responses|chat	Specify the upstream API format, default auto auto-detect
--bridge	Enable automatic 3×3 interconversion among the Anthropic, Responses, and Chat protocols
--url / MOCK_OLLAMA_BASE_URL	Upstream API root address
--apikey / MOCK_OLLAMA_API_KEY	Upstream API Key
--host / --port	Main service listening address and port, default localhost:11434
--quiet	Disable verbose console logging

When the upstream and client protocols differ, --bridgemust be enabled. Common combinations:

# Claude Code → GPT-5.6 / Responses 上游
mock-ollama --url "https://api.example.com/v1" --apikey "你的上游密钥" --api-style responses --bridge

# Codex / GPT-5.6 客户端 → Claude / Anthropic 上游
mock-ollama --url "https://api.anthropic.com" --apikey "你的上游密钥" --api-style anthropic --bridge

The bridge supports text, SSE, tool calls, system messages, and reasoning; when disabled, each protocol is transparently forwarded along the original path and with the original request body.

Claude Code

Claude Code uses the Anthropic Messages route. If the upstream is not Anthropic, the main service must include --bridge。

olm_claude() {
    local model="${1:-claude-sonnet-4-6}"
    shift || true
    ANTHROPIC_AUTH_TOKEN=ollama \
    ANTHROPIC_BASE_URL=http://localhost:11434 \
    CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 \
    ANTHROPIC_MODEL="$model" \
    ANTHROPIC_DEFAULT_SONNET_MODEL="$model" \
    ANTHROPIC_DEFAULT_OPUS_MODEL="$model" \
    ANTHROPIC_DEFAULT_HAIKU_MODEL="$model" \
    claude --permission-mode acceptEdits --model "$model" "$@"
}

# 例：通过 Responses 上游运行 Claude Code
olm_claude claude-sonnet-4-6

Codex and GPT-5.6

Codex uses the OpenAI Responses route. Create a temporary configuration file without modifying the existing ~/.codex/config.toml：

# /tmp/mock-ollama-codex.toml
model_provider = "mock-ollama"
model = "gpt-5.6-terra"

[model_providers.mock-ollama]
name = "mock-ollama"
base_url = "http://localhost:11434/v1"
env_key = "MOCK_OLLAMA_CODEX_API_KEY"
wire_api = "responses"
requires_openai_auth = false

At startup, provide any non-empty local Bearer value; the main service uses its own MOCK_OLLAMA_API_KEY to access the real upstream:

mkdir -p /tmp/mock-ollama-codex-home
cp /tmp/mock-ollama-codex.toml /tmp/mock-ollama-codex-home/config.toml
MOCK_OLLAMA_CODEX_API_KEY=ollama \
  CODEX_HOME=/tmp/mock-ollama-codex-home \
  codex --model gpt-5.6-terra

If the upstream is not Responses, add --bridgewhen starting the main service. You can replace gpt-5.6-terra with the actual upstream model name.

Cursor BYOK

Do not expose 11434 with management interfaces directly to the public network. Cursor uses an independent minimal public network entry point:

# 1. 主服务只留在本机
mock-ollama \
  --url "https://api.example.com/v1" \
  --apikey "你的上游密钥" \
  --api-style responses \
  --bridge

# 2. Cursor 专用入口，默认启动临时 Cloudflare Tunnel
mock-ollama --cursor

The second command prints the configuration required for Cursor Settings → Models:

• Open Override OpenAI Base URL
• Fill in the printed Base URL
• Fill in the printed OpenAI API Key
• Add any of the printed model names; multiple models can be added for free switching

Parameter	Default value	Description
--cursor	-	Start the Cursor BYOK dedicated entry point
--cursor-port	11435	Cursor local loopback port
--cursor-api-key	Randomly generated	Public network entry Bearer key
--cursor-upstream	http://localhost:11434	Already-started main service
--cursor-tunnel	quick	quick Start a temporary Tunnel,off only start the local entry point

The Cursor entry point only exposes /healthz、/v1/models、/v1/chat/completions; except for health checks, all require Bearer authentication.trycloudflare.com The temporary address changes on every restart; for long-term use, a controlled domain and Cloudflare Access should be configured.

Check the service

curl http://localhost:11434/api/version
curl http://localhost:11434/api/tags

curl -X POST http://localhost:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "你的模型名",
    "messages": [{"role": "user", "content": "你好"}]
  }'