Connect claude-code-action with the LLM Gateway

[datvotien231295]

Hi team, I need some advice on this case.
Currently, my company wants to use Claude for code review in pull request. However, according to the integration documentation, the GitHub Action is calling directly to the api.anthropic domain.
I’d like to ask if there’s a way to use it through our company’s LLM Gateway instead.
Is this approach feasible?

  - name: Automatic PR Review
    uses: anthropics/claude-code-action@beta
    env:
      ANTHROPIC_BASE_URL: ${{ secrets.LLM_GATEWAY_BASE_URL }} // domain of LLM internal
    with:
      anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }} //key generate by LLM tools

If you need any further information, feel free to ask me.

[tomoish]

@datvotien231295

Yes, you can run Claude Code Action through your own LLM Gateway instead of hitting api.anthropic.com directry.

You only need to override two parameters:
ANTHROPIC_BASE_URL: ${{ secrets.LLM_GATEWAY_BASE_URL }}
anthropic_api_key: ${{ secrets.LLM_GATEWAY_API_KEY }}

[uc4w6c]

I agree with tomoish’s answer, and I’ll also share the official documentation and an example workflow that might be helpful.

Official documentation: https://docs.anthropic.com/ja/docs/claude-code/llm-gateway

Example workflow:

      - name: LiteLLM run
        run: docker run -d -e ANTHROPIC_API_KEY -p 4000:4000 ghcr.io/berriai/litellm:main-latest --model claude-sonnet-4-20250514
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
      - name: Code Review with Claude
        uses: anthropics/claude-code-action@beta
        with:
          mode: experimental-review
          anthropic_api_key: sk-1234
          custom_instructions: |
            Focus on:
            - Code quality and maintainability
            - Security vulnerabilities
            - Performance issues
            - Best practices and design patterns
            - Test coverage gaps

            Be constructive and provide specific suggestions for improvements.
            Use GitHub's suggestion format when proposing code changes.
        env:
          ANTHROPIC_BASE_URL: http://localhost:4000

[Amansaxena001]

does this work as I tried with portkey and the claude code was not able to authenticate, it was stuck @uc4w6c

[uc4w6c]

@Amansaxena001
It seems to work with Portkey as well.

      - name: Portkey run
        run: npx @portkey-ai/gateway &

      - name: Run Claude Code
        id: claude
        uses: anthropics/claude-code-action@v1
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          claude_args: |
            --allowedTools "Bash(bun install),Bash(bun test:*),Bash(bun run format),Bash(bun typecheck)"
            --model "claude-opus-4-1-20250805"
        env:
          ANTHROPIC_BASE_URL: http://localhost:8787
          ANTHROPIC_AUTH_TOKEN: ${{ secrets.ANTHROPIC_API_KEY }}
          ANTHROPIC_CUSTOM_HEADERS: "x-portkey-provider: anthropic"

[Amansaxena001]

@uc4w6c you would ve to pass the portkey key right not the anthropic one as if we pass the original key then no point of using port key

[uc4w6c]

@Amansaxena001
Then maybe you can just add the API key to ANTHROPIC_CUSTOM_HEADERS, right?
I'm not very familiar with Portkey, so I think it would be better to confirm this on the Portkey team.

[Amansaxena001]

Also have you used custom mcp's like sentry in claude code action? My setup seems right but still when using V1 action we are not able to see connected , the allowedTools are also not present in the init response

{
  "type": "system",
  "subtype": "init",
  "cwd": "/home/runner/work/krypton-ui/krypton-ui",
  "session_id": "100fa2c5-913a-40e3-a088-d39cc4c4bd22",
  "tools": [
    "Task",
    "Bash",
    "Glob",
    "Grep",
    "ExitPlanMode",
    "Read",
    "Edit",
    "Write",
    "NotebookEdit",
    "WebFetch",
    "TodoWrite",
    "WebSearch",
    "BashOutput",
    "KillShell",
    "Skill",
    "SlashCommand"
  ],
  "mcp_servers": [],
  "model": "claude-sonnet-4-5-20250929",
  "permissionMode": "default",
  "slash_commands": [
    "compact",
    "context",
    "cost",
    "init",
    "output-style:new",
    "pr-comments",
    "release-notes",
    "todos",
    "review",
    "security-review"
  ],
  "apiKeySource": "ANTHROPIC_API_KEY",
  "claude_code_version": "2.0.25",
  "output_style": "default",
  "agents": [
    "general-purpose",
    "statusline-setup",
    "output-style-setup",
    "Explore"
  ],
  "skills": [],
  "uuid": "d098ca2d-efe8-48bf-87f5-825481422da5"
}
{
  "type": "assistant",
  "message": {
    "model": "claude-sonnet-4-5-20250929",
    "id": "msg_01KecmVbhmWZEddgjWW34Zv3",
    "type": "message",
    "role": "assistant",

Setup

          claude_args: |
            --model claude-sonnet-4-5-20250929
            --allowedTools mcp__sentry__*,mcp__sentry__get_sentry_issue,Read,View,Bash,Edit,Create,Write,str_replace_editor
            --mcp-config '{
              "mcpServers": {
                "sentry": {
                  "command": "pipx",
                  "args": ["run", "mcp-server-sentry", "--auth-token", "${{ secrets.SENTRY_AUTH_TOKEN }}"],
                  "env": {
                    "SENTRY_AUTH_TOKEN": "${{ secrets.SENTRY_AUTH_TOKEN }}"
                  }
                }
              }
            }'

Though this mcp setup works fine in beta version of the action. I really appreciate your help here