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:

Variable	Where to get it
`ANTHROPIC_API_KEY`	console.anthropic.com → API Keys
`GITHUB_PAT` / `GITLAB_PAT` / `BITBUCKET_PAT`	Your provider's token settings (see setup guides)

Step 2 · Pick your provider and copy the pipeline

Click the section for your git host:

🐙 GitHub

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

🦊 GitLab (cloud)

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

🦊 GitLab (self-hosted)

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

🪣 Bitbucket Cloud

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

🏢 Bitbucket Server / Data Center

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

In Azure DevOps → Pipelines → New pipeline
Connect to your repository
Choose Existing YAML file and point to your file
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:

Action	Input	What happens
`reviewPR`	PR number	Fetches the diff, sends to Claude, posts structured review comment
`createPR`	Source + target branch	Opens a PR on your git host; optionally runs AI review immediately
`commentPR`	PR number + text	Posts a custom comment (build results, status, notes)

AI providers

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

Provider	`aiProvider` value	Auth needed
Anthropic (direct)	`anthropic`	API key from console.anthropic.com
Azure AI Foundry	`azure`	Azure deployment API key + endpoint URL
LiteLLM	`litellm`	Your proxy URL (API key optional)
AWS Bedrock	`bedrock`	IAM role or AWS access key + secret + region
Google Vertex AI	`vertex`	GCP project ID + region + Application Default Credentials

Anthropic (direct API) — default

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

Azure AI Foundry

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

AWS Bedrock

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).

Google Vertex AI

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

LiteLLM (self-hosted proxy)

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):

Model	Speed	Cost (Anthropic)	Best for
`claude-haiku-4-5-20251001`	⚡ Fast	$0.80 / 1M tokens	High volume, quick feedback
`claude-sonnet-4-6`	◎ Balanced	$3 / 1M tokens	General use — recommended
`claude-opus-4-6`	◎ Thorough	$15 / 1M tokens	Security audits, complex changes

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:

Variable	Value
`PrUrl`	Full URL to the PR
`PrNumber`	Numeric PR ID
`ReviewVerdict`	`lgtm` · `needs-work` · `critical`
`ReviewTotalIssues`	Number of issues found
`ReviewSummary`	One-line summary from Claude

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

Supported providers

Provider	Create PR	AI Review	Comment	Self-hosted
GitHub	✅	✅	✅	GitHub Enterprise ✅
GitLab	✅	✅	✅	✅ via `serverUrl`
Bitbucket Cloud	✅	✅	✅	—
Bitbucket Server / DC	✅	✅	✅	✅ via `serverUrl`

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