Skip to main content
Glama

Datacircle tools for the Claude Agent SDK

Find the best GTM APIs. We benchmark all the GTM APIs and we put the results on our website.

Our website: https://datacircle.dev

This package gives the Claude Agent SDK four tools that call Datacircle's open API, https://api.datacircle.dev/v1. The SDK's createSdkMcpServer() puts the tools on an in-process MCP server. query() gives them to the model. You need no key and no account.

Tool

What the tool gives

Call

list_apis

Every GTM API on Datacircle, with the API's slug, number of plans, number of providers and best plan

GET /apis

best_plans

One API's plans, best score out of 100 first. Each plan comes with a rank and a price per 1,000 calls

GET /apis/{slug}/best

search_plans

Plans that match a provider or plan name, on all APIs or on one API. You can sort them by score, provider or price

GET /plans

provider_ranks

The APIs a provider sells, and the provider's rank on each API

GET /providers/{slug}

Install

npm install @datacircle/claude-agent-sdk @anthropic-ai/claude-agent-sdk zod

The tools work with @anthropic-ai/claude-agent-sdk 0.2.0 and later, and zod 4. The package is ESM, like the SDK.

Related MCP server: CLIranks

Use the tools

Put the tools on a server, and give the server to query():

import { createSdkMcpServer, query } from '@anthropic-ai/claude-agent-sdk';
import { datacircleTools } from '@datacircle/claude-agent-sdk';

const datacircle = createSdkMcpServer({ name: 'datacircle', tools: datacircleTools() });

for await (const message of query({
  prompt: 'Who are the 3 best providers for the email-finder API?',
  options: {
    mcpServers: { datacircle },
    allowedTools: ['mcp__datacircle__*'],
  },
})) {
  if (message.type === 'result' && message.subtype === 'success') console.log(message.result);
}

The model sees each tool as mcp__datacircle__<name>, like mcp__datacircle__best_plans. The datacircle in that name is the server's key in mcpServers.

allowedTools lets the tools run with no permission prompt. Leave out allowedTools and canUseTool, and the SDK refuses each call.

Recent versions of the SDK hide MCP tools from the model. The model finds the tools with the SDK's tool search. To show the model all four Datacircle tools from the start, make the server with alwaysLoad: true.

You can put the tools beside your own tools on the same server: tools: [...datacircleTools(), myTool]. To give the model only some of the tools, filter them by name: datacircleTools().filter((t) => t.name !== 'provider_ranks'). A subagent can call the tools too.

Make each call yourself

You can make each call without a model. Each call returns the answer as an object:

import { bestPlans } from '@datacircle/claude-agent-sdk';

const answer = await bestPlans({ slug: 'email-finder', limit: 3 });
console.log(answer.plans);

The other calls are listApis(), searchPlans() and providerRanks().

What the model reads

Each tool gives the model a JSON object as text. Each plan comes with a rank, provider, plan, score, price per 1,000 (perThousand, in dollars) and a page on datacircle.dev. The page on datacircle.dev lists every field.

An unknown slug, or a call with no data on the API, gets an error (isError) with the API's message. If an input does not fit the schema, the tool returns an error that names each wrong argument. If the network fails, the model gets an error (isError) with the network error's message. After an error, the model can call the tool again and keep working.

interrupt() stops the run and the tool's call. On SDK 0.2.141 and later, the run's abortController does the same.

datacircleTools({ fetch }) sets the fetch the tools use, for example to add a proxy or a timeout. To call a tool from your code without a model, pass { fetch } as the call's second argument, like bestPlans({ slug: 'serp' }, { fetch }).

Network

Each tool call makes one HTTPS GET request to https://api.datacircle.dev/v1 and nothing else. The request holds only the tool's own arguments: no key and nothing from your chats. The API takes up to 60 calls a minute from one IP address. After a 429 answer (too many calls), the tool waits up to a minute and tries once more. The package depends only on the SDK and zod. The package stores nothing.

MIT license.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      The API layer for AI agents. World's biggest API index with 22,000+ APIs and growing. Agents discover and call APIs at runtime with semantic search, structured metadata, and 18 Direct Call APIs including AI providers.
      14
      616 npm
      8
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      CLIRank - API discovery for AI agents. APIs scored on CLI relevance: auth method, JSON responses, headless operation, pricing transparency. Search by capability, compare alternatives, read agent-contributed docs. 6 MCP tools for discovery, comparison, and documentation. Agents contribute observations back - crowdsourced API docs by machines, for machines.
      8
      69 npm
      6
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables Claude to discover and call any REST API from an OpenAPI/Swagger spec or Postman collection at runtime, with authentication, workflows, and prebuilt registry support.
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables agents to rank model variants and hunt down API offers for a specific task, estimate real billed cost, audit n8n/Make automations for overpay, missing fallbacks and deprecated models, and plan or run evaluation and canary suites on the user's own examples. It also supports price comparison, custom reseller channels with measured per-call cost, and non-LLM capabilities such as web search and speech-to-text.
      AGPL 3.0