Integration claude-agent
Give Claude Agent the web without giving up your privacy.
Cited answers, clean Markdown, matching JSON, and browser tasks for the Claude Agent SDK, as native Claude Agent SDK tools.
@tabstack/claude-agent
ollama.com/blog
docs.openwebui.com
ANSWER
Two runtimes shipped it.
Why Claude Agent
Tabstack, inside Claude Agent.
@tabstack/claude-agent registers Tabstack as in-process tools for the Claude Agent SDK. Your agent gets cited research, clean page content, matching JSON, and browser tasks inside query(), with no separate server to run.
post /research
query: "which runtimes…"
mode: "balanced"First call
Start with a current question.
Claude Agent makes the call, and Tabstack handles the web. Research searches live sources, reads the pages, and returns a cited answer, so your agent gets the findings without running every search itself.
npm install @tabstack/claude-agent @anthropic-ai/claude-agent-sdk zod
export TABSTACK_API_KEY="your-tabstack-key"
export ANTHROPIC_API_KEY="your-anthropic-key"
import { query } from "@anthropic-ai/claude-agent-sdk";
import { tabstackServer, tabstackAllowedTools, tabstackMcpServerName } from "@tabstack/claude-agent";
for await (const message of query({
prompt: "What are Vercel's current pricing plans, with sources?",
options: {
mcpServers: { [tabstackMcpServerName]: tabstackServer },
allowedTools: tabstackAllowedTools,
},
})) {
if (message.type === "result" && message.subtype === "success") {
console.log(message.result);
}
}
Tools
Five tools, one key.
Each Tabstack endpoint is a tool your agent can call. tabstackAllowedTools pre-approves all five, or list only the ones a run should reach.
| Tool | Returns |
|---|---|
research_question | A cited answer from multiple live sources |
extract_page_content | A page as clean Markdown |
extract_structured_data | JSON matching a schema you define |
generate_structured_data | A page transformed into the JSON you describe |
automate_browser_task | A completed browser task, from plain-language steps |