AI PR Reviewer

Automatic AI code reviews on GitHub, GitLab & Bitbucket — triggered directly from Azure DevOps pipelines.

No extra servers. No webhooks to configure. One task, three providers, real Claude reviews posted as PR comments.

Quick Start — 3 steps

Step 1  ·  Store your secrets

In Azure DevOps → Pipelines → Library → Variable groups, create a group (e.g. ai-reviewer-secrets) with two secret variables:

Step 2  ·  Pick your provider and copy the pipeline

Click the section for your git host:

trigger: none

pr:
  branches:
    include: [main]

variables:
- group: ai-reviewer-secrets   # contains GITHUB_PAT + ANTHROPIC_API_KEY

pool:
  vmImage: ubuntu-latest

steps:
- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    provider: github
    accessToken: $(GITHUB_PAT)
    repository: myorg/myrepo          # ← change this
    prNumber: $(System.PullRequest.PullRequestNumber)
    enableAiReview: true
    aiApiKey: $(ANTHROPIC_API_KEY)
    aiModel: claude-sonnet-4-6

trigger: none

pr:
  branches:
    include: [main]

variables:
- group: ai-reviewer-secrets   # contains GITLAB_PAT + ANTHROPIC_API_KEY

pool:
  vmImage: ubuntu-latest

steps:
- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    provider: gitlab
    accessToken: $(GITLAB_PAT)
    repository: mygroup/myproject     # ← change this
    prNumber: $(System.PullRequest.PullRequestNumber)
    enableAiReview: true
    aiApiKey: $(ANTHROPIC_API_KEY)
    aiModel: claude-sonnet-4-6

trigger: none

pr:
  branches:
    include: [main]

variables:
- group: ai-reviewer-secrets   # contains GITLAB_PAT + ANTHROPIC_API_KEY

pool:
  vmImage: ubuntu-latest

steps:
- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    provider: gitlab
    accessToken: $(GITLAB_PAT)
    repository: mygroup/myproject             # ← change this
    serverUrl: https://gitlab.mycompany.com   # ← change this
    prNumber: $(System.PullRequest.PullRequestNumber)
    enableAiReview: true
    aiApiKey: $(ANTHROPIC_API_KEY)
    aiModel: claude-sonnet-4-6

trigger: none

pr:
  branches:
    include: [main]

variables:
- group: ai-reviewer-secrets   # contains BITBUCKET_USERNAME + BITBUCKET_APP_PASSWORD + ANTHROPIC_API_KEY

pool:
  vmImage: ubuntu-latest

steps:
- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    provider: bitbucket
    accessToken: $(BITBUCKET_USERNAME):$(BITBUCKET_APP_PASSWORD)   # format: username:app_password
    repository: myworkspace/myrepo           # ← change this
    prNumber: $(System.PullRequest.PullRequestNumber)
    enableAiReview: true
    aiApiKey: $(ANTHROPIC_API_KEY)
    aiModel: claude-sonnet-4-6

Bitbucket App Password: Personal settings → App passwords → Create → enable Repositories: Read and Pull requests: Read & Write

trigger: none

pr:
  branches:
    include: [main]

variables:
- group: ai-reviewer-secrets   # contains BITBUCKET_PAT + ANTHROPIC_API_KEY

pool:
  vmImage: ubuntu-latest

steps:
- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    provider: bitbucket-server
    accessToken: $(BITBUCKET_PAT)
    repository: MYPROJECT/myrepo              # ← PROJECT_KEY/repo-slug
    serverUrl: https://bitbucket.mycompany.com  # ← change this
    prNumber: $(System.PullRequest.PullRequestNumber)
    enableAiReview: true
    aiApiKey: $(ANTHROPIC_API_KEY)
    aiModel: claude-sonnet-4-6

Step 3  ·  Create the ADO pipeline

1. In Azure DevOps → Pipelines → New pipeline
2. Connect to your repository
3. Choose Existing YAML file and point to your file
4. Open a pull request → the pipeline runs and posts the AI review

What it does

The task has three actions you can mix and match:

AI providers

Choose where to run the model. All providers host Claude models; the only difference is authentication and model naming.

- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    # ... git provider inputs ...
    enableAiReview: true
    aiProvider: anthropic
    aiApiKey: $(ANTHROPIC_API_KEY)
    aiModel: claude-sonnet-4-6

Requires a Claude deployment in Azure AI Foundry. Find the endpoint URL in Azure AI Foundry → Project → Deployments → your model → API endpoint.

- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    # ... git provider inputs ...
    enableAiReview: true
    aiProvider: azure
    aiApiKey: $(AZURE_AI_API_KEY)
    aiBaseUrl: $(AZURE_AI_ENDPOINT)      # e.g. https://<resource>.services.ai.azure.com/models
    aiModel: claude-sonnet-4-6           # your deployment name

Enable the Claude model in AWS Console → Bedrock → Model access for your target region. Uses IAM role credentials automatically; supply keys only if not running on an EC2/ECS role.

- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    # ... git provider inputs ...
    enableAiReview: true
    aiProvider: bedrock
    awsRegion: us-east-1
    awsAccessKeyId: $(AWS_ACCESS_KEY_ID)         # optional — leave blank to use IAM role
    awsSecretAccessKey: $(AWS_SECRET_ACCESS_KEY) # optional — leave blank to use IAM role
    aiModel: anthropic.claude-3-5-sonnet-20241022-v2:0

Bedrock model IDs use the anthropic. prefix. Newer models on cross-region inference use the us. / eu. prefix (e.g. us.anthropic.claude-opus-4-5:0).

Enable the Claude model in Google Cloud Console → Vertex AI → Model Garden. Authentication uses Application Default Credentials — set GOOGLE_APPLICATION_CREDENTIALS to your service account key file path in the pipeline.

- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    # ... git provider inputs ...
    enableAiReview: true
    aiProvider: vertex
    gcpProjectId: my-gcp-project
    gcpRegion: us-east5
    aiModel: claude-sonnet-4-6

Point the task at your LiteLLM proxy. The proxy handles the actual model routing. API key is optional depending on your proxy configuration.

- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    # ... git provider inputs ...
    enableAiReview: true
    aiProvider: litellm
    aiBaseUrl: http://litellm.internal:4000   # your proxy URL
    aiApiKey: $(LITELLM_API_KEY)              # optional
    aiModel: claude-sonnet-4-6

Choose a model

All providers support these Claude models (use the exact ID for your provider):

AWS Bedrock uses different model IDs — see the Bedrock section above. Pricing on Azure / Bedrock / Vertex may differ from Anthropic direct.

Typical cost per PR: $0.001 – $0.05 depending on diff size and model.

Customize the review focus

Use aiReviewContext to tell Claude what to prioritize:

- task: AiPrReviewer@1
  inputs:
    action: reviewPR
    provider: github
    accessToken: $(GITHUB_PAT)
    repository: myorg/myrepo
    prNumber: $(System.PullRequest.PullRequestNumber)
    enableAiReview: true
    aiApiKey: $(ANTHROPIC_API_KEY)
    aiModel: claude-sonnet-4-6
    aiReviewContext: |
      Focus on security vulnerabilities and breaking API changes.
      Flag any hardcoded credentials or exposed secrets.
    aiMaxDiffLines: 500   # truncate large diffs (default 500)

Output variables

After reviewPR or createPR, downstream steps can read:

- script: |
    echo "PR: $(PrUrl)"
    echo "Verdict: $(ReviewVerdict) — $(ReviewTotalIssues) issues"

Supported providers

Documentation

• GitHub setup
• GitLab setup
• Bitbucket setup
• Azure DevOps setup
• User guide — all actions, models, cost estimates
• Troubleshooting
• FAQ
• Source code

License: MIT  ·  Publisher: subzone  ·  Powered by Anthropic Claude