@geekflare/ai-sdk package gives the Vercel AI SDK Geekflare as a set of tools: web and Google SERP search, page scraping to Markdown or structured JSON, screenshots, and website diagnostics. It works with AI SDK 5, 6, and 7, requires zod 3.25+ or 4 and Node 18+, supports both ESM and CJS, and has no dependencies beyond the ai and zod peers.
Install
Set the API key
Get an API key at dash.geekflare.com and set it as an environment variable:YOUR_API_KEY with your Geekflare API key. You can also pass the key to any tool with the apiKey option.
Manage the key with Vercel Connect
On Vercel, you can keep the key out of your code and environment variables with Vercel Connect. You add the key to a Geekflare connector once, and your app fetches it at runtime. From your project folder, link the project and create the connector. The CLI asks for your Geekflare API key:vercel env pull writes a VERCEL_OIDC_TOKEN to .env.local so getToken also works locally. Deployed functions get this token automatically.
Install the Connect SDK, fetch the key, and pass it to the tools with apiKey:
vercel connect attach from that project.
Quick start
AddgeekflareSearch() to generateText so the model can search the web:
All tools
geekflareTools() returns all four tools, keyed by name:
Options
Every tool accepts these options:
Some tools also take defaults that apply when the model doesn’t set them:
Errors
API errors (such as an invalid key, no credits, or an unreachable site) throwGeekflareAPIError with the message from the API. The AI SDK returns it to the model as a tool error, so the model can read the message and respond.
Notes
- Each call spends credits from your Geekflare account. See Credit Consumption and pricing.
- Geekflare works with public pages only.
Links
npm
@geekflare/ai-sdk on npm.GitHub
Source code (MIT license).
LangChain
Use Geekflare tools in LangChain agents.
LlamaIndex
Use Geekflare tools in LlamaIndex agents.
MCP Server
Connect Geekflare to AI clients through MCP.