Agent Tools / For agent builders
Free tools

For agent builders MCP + HTTP · no key

Live public data for your agent, in one line

One MCP address and no key: job boards, app reviews, YouTube transcripts and X posts come back as structured results with the source and the time they were read. The index tool finds the rest when your agent needs it.

01 Pick your assistant

Works with

02 Add the server, then ask

Claude Code

claude mcp add --transport http openkrill https://api.openkrill.app/mcp

Run it in a terminal, then ask. Check: claude mcp get openkrill.

Cursor

{
  "mcpServers": {
    "openkrill": {
      "url": "https://api.openkrill.app/mcp"
    }
  }
}

The button opens Cursor to confirm. Or add this to ~/.cursor/mcp.json.

ChatGPT

https://api.openkrill.app/mcp

Settings, Developer mode on. Plugins, plus, paste under Connection.

Claude

https://api.openkrill.app/mcp

Customize, Connectors, Add custom connector, paste the address.

n8n

https://api.openkrill.app/mcp

AI Agent node, MCP Client Tool: this Endpoint, HTTP Streamable, no auth.

Add to Claude Code Free, about 1 minute, nothing to install.

Your agent calls list_jobs({ companies: ["greenhouse:vercel"], remote: true, postedSince: "2026-09-01" })

list_jobs answers

3 open roles at Vercel posted since 1 September, each with its board link and posting date:

  • 24 Sep 2026 Product Manager, Dashboard Product, San Francisco.
  • 16 Sep 2026 Software Engineer, Data Platform Data, New York.
  • 15 Sep 2026 Design Engineer Design, San Francisco.

Read from Vercel's Greenhouse board on 15 October 2026. The same call works over plain HTTP.

Try a tool live · Every tool

KrillData: an MCP server and REST API that gives AI agents web and social data.

Try the 4 most used tools

Real calls, real answers. Edit and run.

Post Kit

513 calls, week of

One topic in, a checked post for each network out.

draft_posts

Page Fetch

77 calls, week of

Any page as clean Markdown, with its links.

fetch_page

Site Check

63 calls, week of

Which AI crawlers your robots.txt lets in.

check_ai_crawler_access

SEO Audit

57 calls, week of

One page, its SEO findings by severity.

audit_page

Find a tool

39 tools by job. Search, or open a group.

Everyday checks 6

Car Check

Decode a VIN, recalls, fuel use

Food Facts

Ingredients and allergens from a barcode

Shops and sellers 6

CertPacket

US safety certificates for imports

Code and security 2
Websites, domains and email 6

SEO Audit

A ranked list of SEO problems

Research and public facts 5

Belief Poll

What people believe, with linked posts

Agent Tools API

Open jobs, app reviews, X posts

Jobs, sales and planning 5

BillPeak

Check a business electricity bill

Writing, media and files 9

Post Stats

Views and likes of any public post

Post Kit

Social posts checked per platform

Send Out

Email drafts, events and share links

Video Grab

Freely licensed videos with their license

Market numbers 1

Information only: no advice, no signals and no trading.

Connect once

https://recalls.openkrill.app/mcp

Install guides: ChatGPT, Claude, Cursor, Pi. Zapier: 40 actions.

Agent Tools weekly: what we measured in the week's posts, and what you can try. At most one email a week.

You confirm by a link we email you. No tracking. Unsubscribe in one click. Past issues. Privacy.

  • 40tool servers
  • 147callable tools
  • 175install guides
  • 0keys to start

Each tool server is free within 60 calls a minute per network and 25000 calls a day per server, shared by all callers; the shared API allows 20 requests a minute and 200 tool calls a day per network. What it costs. How it compares.

X posts in bulk: X scraper tutorial. X scrapers compared.

Our own test, checked 9 October 2026: on 20 questions about new CVEs, end-of-life dates and paper citations, a model alone got 14 right; with our servers it got 20 right, each graded against its primary source. Every question and grade.

For agents

Text for agents: llms.txt, index.md, llms-full.txt and MCP at api.openkrill.app/mcp.

Frequently asked questions

What is Agent Tools and how do AI assistants use it?

Agent Tools gives AI agents and developers live public data over HTTP and Model Context Protocol (MCP). An assistant connects to the MCP server endpoint and calls tools for job listings, app reviews, domain checks, security advisories, and public records.

Do I need an API key or an account to use these tools?

No. You can start without an account or an API key. Requests without a key share a network allowance and have rate limits. An API key is optional, and if you send one in the x-api-key header, those limits are lifted. To get a key, email support@openkrill.app.

What are the anonymous rate limits and caching policies?

Without an API key, the shared API at api.openkrill.app takes 20 requests per minute and 200 tool calls per day per network. Responses are cached at edge nodes, and the cache status comes back as HIT, MISS, or STALE. A repeat request is served straight from cache.

Which AI clients are supported?

Any client that supports Model Context Protocol streamable HTTP can connect, including Claude, Cursor, ChatGPT, and Pi. Each one has install instructions and configuration examples.

What is the difference between the HTTP API and the MCP server?

The HTTP API returns standard JSON over REST endpoints at api.openkrill.app. The MCP server speaks the Model Context Protocol over streamable HTTP at api.openkrill.app/mcp. It exposes tool schemas and runs tools/call requests directly inside AI workflows.

How much does Agent Tools cost?

The tools are free within the rate and daily limits, with no sign-up. Stillchat, our iPhone app, is free too. Paid help is a done-for-you service at a fixed price, and every price is on the prices page.

What if no tool does what I need?

Ask from inside your assistant. Every server has a submit_feedback tool: send the kind need_tool with what you wanted, then call get_feedback_reply with the ticket to read our reply. You can also email support@openkrill.app.

Why did a tool answer rate_limited or unavailable?

rate_limited means a limit was reached. That's either ours (60 calls a minute per network on each tool server) or the limit of the public source the tool reads. Wait a minute and ask again. unavailable means the source didn't answer, so the tool says so instead of guessing. Neither one means there are no results.

How do I report a wrong answer?

Call submit_feedback with the kind bug, the tool name, what you asked, what came back, and what is right. Each report gets a ticket, and get_feedback_reply shows you the reply.

What do the tools keep about my requests?

Each tool sends only its arguments to the public source it reads. It keeps daily call counts, not your conversation. A few tools keep a file or a snapshot you asked for, for a set time. The privacy page lists every server, what it sends where, and what it keeps.