Run Claude Code on Ollama or DeepSeek — No Anthropic Bill Required

Personal use case: The claude model is really too expensive. Since Claude Code natively only supports the Anthropic API format, this article documents how to connect local models or other models (DeepSeek, etc.) to Claude Code.

Via Claude Code Router (CCR): Connect the Claude Code CLI to Ollama (local models) or DeepSeek (cloud models), enabling Claude Code to run with non-Anthropic models.

---

Table of Contents

1. Principle Overview
2. Installing Claude Code Router
3. Configuring the Ollama Provider
4. Configuring the DeepSeek Provider
5. Configuring Claude Code to Connect to CCR
6. Verifying the Connection
7. Advanced: Routing Rules and Virtual Models
8. Common Troubleshooting
9. Reference Links

---

1. Principle Overview

Architecture Diagram

┌─────────────────┐      Anthropic Messages API       ┌──────────────────────────┐
│                 │  ──────────────────────────────▶   │                          │
│   Claude Code   │      /v1/messages                 │  Claude Code Router      │
│   (CLI / VSC)   │  ◀──────────────────────────────   │  (localhost:3456)        │
│                 │                                    │                          │
└─────────────────┘                                    └──────────┬───────────────┘
                                                                  │
                           ┌──────────────────────────────────────┼──────────────────────────┐
                           │                                      │                          │
                           ▼                                      ▼                          ▼
                  ┌──────────────────┐                  ┌──────────────────┐      ┌──────────────────┐
                  │   Ollama         │                  │   DeepSeek API   │      │   OpenAI API     │
                  │   (localhost)    │                  │   (Cloud)        │      │   (Other Compat) │
                  │   :11434/v1      │                  │   api.deepseek   │      │                  │
                  └──────────────────┘                  └──────────────────┘      └──────────────────┘

Workflow

1. Claude Code natively only supports the Anthropic Messages API format (/v1/messages)
2. Claude Code Router runs locally, exposing an Anthropic-compatible endpoint at http://127.0.0.1:3456
3. By setting ANTHROPIC_BASE_URL=http://127.0.0.1:3456, Claude Code sends all API requests to CCR
4. CCR performs protocol conversion internally: Translates the Anthropic Messages format into the OpenAI Chat Completions format (supported by both Ollama and DeepSeek)
5. The response is then translated back from the OpenAI format to the Anthropic format and returned to Claude Code

Key Facts

Item	Description
Protocol sent by Claude Code	Anthropic Messages API (/v1/messages)
Protocol supported by Ollama	OpenAI Chat Completions API (/v1/chat/completions)
Protocol supported by DeepSeek	OpenAI Chat Completions API (/v1/chat/completions)
Role of CCR	Protocol translation gateway, translating between the two formats
Authentication method	Credentials passed via ANTHROPIC_AUTH_TOKEN or ANTHROPIC_API_KEY

---

2. Installing Claude Code Router

System Requirements

• macOS 10.15+ / Windows 10+ / Linux (supports AppImage)
• Claude Code CLI installed

Download and Install

1. Visit the GitHub Releases page
2. Download the installation package for your system:
   • macOS: .dmg file
   • Windows: .exe installer
   • Linux: .AppImage file
3. Install and launch the application

First Launch

After launching, CCR creates a configuration file at:

• macOS / Linux: ~/.claude-code-router/config.json
• Windows: %APPDATA%\Claude Code Router\config.json

---

3. Configuring the Ollama Provider

Prerequisites

• Ollama installed
• At least one model pulled, for example:

ollama pull llama3.1
ollama pull qwen2.5
ollama pull deepseek-r1:7b

• The Ollama service is running (default listening on http://127.0.0.1:11434)

Configuration via CCR Interface

1. Open the CCR desktop application
2. Go to Providers → Click Add Provider
3. Fill in the following information:

Field	Value
Provider Name	Ollama
Endpoint (Base URL)	http://127.0.0.1:11434/v1
Protocol	openai_chat_completions
API Key	Leave blank (or fill in ollama arbitrarily)
Models	Enter the names of models you have pulled, e.g., llama3.1, qwen2.5

1. Click Save

Why is the Ollama endpoint :11434/v1? Starting from version 0.1.32, Ollama natively supports an OpenAI-compatible API at the path /v1/chat/completions. CCR uses the openai_chat_completions protocol to convert Anthropic requests and send them to this endpoint.

Optional: Configuration via Deeplink

You can also add it quickly via a URL Scheme:

ccr://provider?name=Ollama&base_url=http%3A%2F%2F127.0.0.1%3A11434%2Fv1&api_key=ollama&models=llama3.1,qwen2.5&protocol=openai_chat_completions

Opening this link in a browser will cause CCR to pop up a confirmation dialog.

Testing the Ollama Connection

Verify that Ollama's OpenAI-compatible API is working correctly from the terminal:

curl http://127.0.0.1:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama3.1",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

It should return a JSON response containing a choices field.

---

4. Configuring the DeepSeek Provider

Prerequisites

• Register on the DeepSeek Platform
• Obtain an API Key (create one in the console)

Configuration via CCR Interface

1. Open the CCR desktop application
2. Go to Providers → Click Add Provider
3. Fill in the following information:

Field	Value
Provider Name	DeepSeek
Endpoint	https://api.deepseek.com/v1
Protocol	openai_chat_completions
API Key	sk-xxxxxxxxxxxxxxxxxxxx (your DeepSeek API Key)
Models	deepseek-chat, deepseek-reasoner

1. Click Save

Optional: Configuration via Deeplink

ccr://provider?name=DeepSeek&base_url=https%3A%2F%2Fapi.deepseek.com%2Fv1&api_key=sk-xxxxxxxxx&models=deepseek-chat,deepseek-reasoner&protocol=openai_chat_completions

DeepSeek Model Description

Model ID	Description	Use Case
deepseek-chat	DeepSeek V3 / Latest chat model	Daily coding, code generation
deepseek-reasoner	DeepSeek R1 reasoning model	Complex reasoning, debugging analysis

---

5. Configuring Claude Code to Connect to CCR

5.1 Start the CCR Gateway

1. In the CCR desktop application, go to Server
2. Click Start to start the gateway service
3. Confirm the status: both endpoints should be running:
   • Wrapper gateway: http://127.0.0.1:3456
   • Core gateway runtime: http://127.0.0.1:3457

💡 It is recommended to check Auto-start, so CCR will automatically start the gateway every time you boot up.

5.2 Create a Claude Code Configuration File

Method A: Global Configuration (Recommended)

Edit ~/.claude/settings.json (macOS/Linux) or %USERPROFILE%\.claude\settings.json (Windows):

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://127.0.0.1:3456",
    "ANTHROPIC_API_KEY": "ccr-local"
  }
}

The value of ANTHROPIC_API_KEY can be any non-empty string — CCR, being a local gateway, does not validate this key, but Claude Code requires a value to bypass the login flow.

Method B: Project-Level Configuration

Create .claude/settings.local.json in the project root directory (this file is automatically gitignored):

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://127.0.0.1:3456",
    "ANTHROPIC_API_KEY": "ccr-local"
  }
}

Method C: Shell Environment Variables (Temporary, Good for Testing)

export ANTHROPIC_BASE_URL=http://127.0.0.1:3456
export ANTHROPIC_API_KEY=ccr-local

5.3 Using CCR Profiles for Automatic Configuration (Recommended)

CCR provides a Profile feature for one-click configuration application:

1. In CCR, go to Profiles
2. Select Claude Code
3. Select the target model (e.g., llama3.1 configured in your Ollama)
4. Click Apply — this automatically sets the environment variables and opens Claude Code

5.4 Configuring CCR Routing

Set the following in the CCR Routing panel:

Routing Item	Recommended Configuration
Default Provider	Select your configured Ollama or DeepSeek
Default Model	Select the specific model, e.g., llama3.1 or deepseek-chat

You can also configure routing rules for different workloads:

• Background (background tasks) → Can use a cheaper model
• Thinking (reasoning tasks) → Can use a more powerful model
• Subagent (sub-agents) → Can use a more cost-effective model

---

6. Verifying the Connection

6.1 Quick Test (curl)

After configuring the environment variables, first test if the CCR gateway responds correctly using curl:

curl -X POST http://127.0.0.1:3456/v1/messages \
  -H "Authorization: Bearer ccr-local" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model": "llama3.1", "max_tokens": 50, "messages": [{"role": "user", "content": "Say hello"}]}'

Note: The model name here must be consistent with the model name you previously added in CCR. For DeepSeek, use deepseek-chat; for Ollama, use the model name you pulled, such as llama3.1.

If it returns JSON containing a content field, CCR is working correctly.

6.2 Launching Claude Code

cd your-project
claude

If everything is normal, Claude Code should skip the login screen and go directly into a session.

6.3 Checking Status

Run the following command in Claude Code:

/status

Check the following key information:

Status Item	Expected Value
Anthropic base URL	http://127.0.0.1:3456
API key	Should display ANTHROPIC_API_KEY (not Login method)

6.4 Sending a Test Message

Try sending a simple message, for example:

Hello, what model are you running on?

If the CCR routing is configured correctly, you will receive a model reply, and you can see the request record in the CCR Dashboard.

---

7. Advanced: Routing Rules and Virtual Models

7.1 Multi-Model Routing

CCR supports routing different tasks to different models based on request characteristics:

Routing Condition	Target Model	Use Case
Default	deepseek-chat	Daily conversation and coding
Background	llama3.1 (local Ollama)	Background tasks, code search
Thinking	deepseek-reasoner	Tasks requiring deep reasoning
Long Context	deepseek-chat	Processing large files

Configuration method: Set it in CCR → Routing → Add Routing Rule.

7.2 Virtual Models

You can create virtual model names to simplify model selection:

llama3.1 → llama3.1 on Ollama
default-coding → deepseek-chat
default-reasoning → deepseek-reasoner

Configure this in CCR → Routing → Virtual Models.

7.3 Fallback Routing

When the primary model is unavailable, CCR can automatically switch to a fallback model:

Primary:   deepseek-chat (DeepSeek)
Fallback:  llama3.1 (Ollama, runs locally, no network required)

Configure this in Routing → Edit routing rule → Configure Fallback model.

7.4 API Key Rotation

For API services like DeepSeek, you can configure multiple API Keys for load balancing and failover:

In Providers → Edit DeepSeek → API Key field, use commas to separate multiple keys:

sk-key1,sk-key2,sk-key3

CCR will automatically rotate to the next key when a request fails.

---

8. Common Troubleshooting

8.1 Claude Code Still Shows the Login Screen

Cause: The CCR gateway configuration is not being read by Claude Code.

Solution:

• Confirm that both ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY are correctly set
• If using settings.json, confirm the file path is correct:
  • Global: ~/.claude/settings.json
  • Project-level: .claude/settings.local.json
• Run echo $ANTHROPIC_BASE_URL to check if the environment variable is effective
• Run /status in Claude Code to see the actual base URL in effect

8.2 Connection Refused

Cause: The CCR gateway is not started.

Solution:

• Confirm the CCR desktop application is running
• In CCR → Server, confirm the Gateway status is "Running"
• Check if the port is occupied: lsof -i :3456

8.3 Model Returns Errors or No Response

Cause: There may be multiple possible issues.

Solution Steps:

For Ollama:

# Check Ollama service status
ollama list

# Confirm the model is loaded
ollama run llama3.1

# Check Ollama logs
ollama serve

For DeepSeek:

# Directly test the DeepSeek API
curl https://api.deepseek.com/v1/chat/completions \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek-chat", "messages": [{"role": "user", "content": "hi"}]}'

8.4 401 Authentication Error

Cause: The API Key is not being passed correctly.

Solution:

• If using ANTHROPIC_API_KEY, CCR receives the x-api-key header
• If using ANTHROPIC_AUTH_TOKEN, CCR receives the Authorization: Bearer header
• In most cases, both work; just ensure the variable name in settings.json matches what CCR expects

8.5 Protocol Conversion Error

Cause: CCR may encounter incompatible fields when converting between Anthropic Messages ↔ OpenAI Chat Completions.

Solution:

• Set the environment variable in Claude Code: CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1
• This causes Claude Code to send fewer experimental fields, reducing the chance of protocol conversion errors
• Add it to settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://127.0.0.1:3456",
    "ANTHROPIC_API_KEY": "ccr-local",
    "CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS": "1"
  }
}

8.6 Checking CCR Logs

The Network Logs panel in the CCR desktop application can view all passing requests and responses, making it the best tool for troubleshooting.

8.7 Resetting Configuration

If you need a complete reset:

# Back up existing configuration
cp ~/.claude-code-router/config.json ~/.claude-code-router/config.json.bak

# Delete configuration (CCR will recreate it on next launch)
rm ~/.claude-code-router/config.json

# Clear Claude Code cache
rm -rf ~/.claude/cache

---

9. Complete Configuration Example

Final ~/.claude/settings.json

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://127.0.0.1:3456",
    "ANTHROPIC_API_KEY": "ccr-local",
    "CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS": "1",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}

Startup Sequence

The correct startup sequence is:

1. Start Ollama (if using local models): ollama serve
2. Start CCR and ensure the gateway is in Running status
3. Start Claude Code: claude

---

10. Reference Links

• Claude Code Router GitHub — Project source code and Release downloads
• Claude Code Official Documentation - LLM Gateway — Claude Code's gateway protocol description
• Claude Code Official Documentation - Connect to LLM Gateway — Detailed guide for configuring ANTHROPIC_BASE_URL
• Claude Code Official Documentation - Gateway Protocol Reference — Gateway API protocol details
• Ollama Official Documentation — Running local models with Ollama
• DeepSeek Platform — DeepSeek API management

---

Disclaimer: Claude Code Router is a third-party open-source project, not officially maintained by Anthropic. When using non-Anthropic models to run Claude Code, some native Claude Code features (such as ultra-long context, thinking capabilities, tool streaming, etc.) may be limited due to protocol conversion. Always use this in compliance with your organization's security policies.