Getting Started
Install @github-tools/sdk, add the peer dependencies for your framework (AI SDK, eve, Vercel Workflow), and configure your GitHub token.

Install @github-tools/sdk in this project

Install the SDK

pnpm add @github-tools/sdk

Pick your peer dependencies

What you install next depends on the framework you build on:

FrameworkImport pathPeer dependencies
TanStack AI@github-tools/sdk/tanstack@tanstack/ai, ai, zod; @tanstack/ai-vercel-gateway for default auto evaluation
AI SDK@github-tools/sdkai (v6 or v7), zod
eve extension@github-tools/eve-extensioneve (declared *, requires 0.81+, ai v7 transitively)
eve (direct import, deprecated)@github-tools/sdk/eveeve, ai v7, zod
Vercel Workflow@github-tools/sdk/workflowworkflow, @ai-sdk/workflow, ai, zod
Chat SDK@github-tools/sdkchat, @chat-adapter/github, ai, zod

Required: AI SDK and Zod

The SDK declares AI SDK and Zod as required package peers. The TanStack entry point uses its own execution loop:

pnpm add ai zod

Optional: TanStack AI

Install @tanstack/ai and a model adapter. The TanStack AI guide includes installation commands for Gateway chat and Jev evaluation. Gateway is optional when you supply another chat adapter and your own evaluation adapter.

For eve filesystem-first agents, install @github-tools/eve-extension and its eve peer (which itself requires ai v7). The extension declares eve as * so your agent supplies the runtime copy; eve checks the extension's generated capability metadata rather than an npm range:

pnpm add @github-tools/eve-extension eve

See Build a GitHub agent with the eve extension: the full agent is 3 files, mounted under agent/extensions/.

Optional: eve direct import (deprecated)

The lower-level @github-tools/sdk/eve import still works for existing agent/tools/ setups, but is deprecated in favor of the extension above. It needs eve and ai v7 on top of the SDK itself:

pnpm add eve ai

See eve (direct import) for the deprecated instructions.

Optional: durable workflows

For durable agents with the Vercel Workflow SDK, add workflow and @ai-sdk/workflow when you import from @github-tools/sdk/workflow and use createDurableGithubAgent:

pnpm add workflow @ai-sdk/workflow

See Durable agents with Vercel Workflow for the full pattern.

Set your GitHub token

The SDK reads GITHUB_TOKEN from your environment automatically. Create a .env file at the root of your project:

.env
GITHUB_TOKEN=github_pat_xxxxxxxxxxxx

You can also pass the token explicitly if you prefer:

explicit-token.ts
createGithubTools({ token: 'github_pat_xxxxxxxxxxxx' })
Never commit tokens. Use .env locally and secret managers (Vercel, GitHub Actions secrets) in CI/production.

Choose a token type

Token typeWhen to useScope control
Fine-grained PATProduction assistantsper-repository, per-permission
Classic PATQuick prototypingbroad, org-level
Vercel ConnectAgents deployed on Vercelper-installation, per-permission, short-lived

See Tokens and authentication for the detailed permission matrix and the Vercel Connect setup.

Verify the setup

Run a minimal script to confirm the SDK initializes and a read tool resolves:

verify.ts
import { createGithubTools } from '@github-tools/sdk'

const tools = createGithubTools({ preset: 'repo-explorer' })

console.log('Tools loaded:', Object.keys(tools).join(', '))

If the script prints tool names, your environment is ready. Continue with Run your first call.

For the packaged skill (same domain), see Agent Skills.