LIVE
OPUS 4.7$15 / $75per Mtok
SONNET 4.6$3 / $15per Mtok
GPT-5.5$10 / $30per Mtok
GEMINI 3.1$3.50 / $10.50per Mtok
SWE-BENCHleader Claude Opus 4.772.1%
MMLU-PROleader Opus 4.788.4
VALS FINANCEleader Opus 4.764.4%
AFTAv1.0 whitepaper live at /whitepaper
OPUS 4.7$15 / $75per Mtok
SONNET 4.6$3 / $15per Mtok
GPT-5.5$10 / $30per Mtok
GEMINI 3.1$3.50 / $10.50per Mtok
SWE-BENCHleader Claude Opus 4.772.1%
MMLU-PROleader Opus 4.788.4
VALS FINANCEleader Opus 4.764.4%
AFTAv1.0 whitepaper live at /whitepaper
All systems operational0 AI providers monitored, polled every 2 minutes
Live status

MCP

the model context protocol

MCP is the open standard for connecting AI agents to external tools, data sources, and APIs. Anthropic published it; OpenAI, Google, Microsoft, and the broader agent ecosystem adopted it. As of May 2026, the official registry has nearly 8,000 servers and is adding hundreds per day. This page is the canonical hub. What MCP is, how the registry is growing, the 50-line pattern for shipping a server, and the latest news in one place.

modelcontextprotocol.io official registry curated catalog/api/mcp/registry/snapshot

TensorFeed ships an MCP server

tensorfeed-mcp exposes 14 paid TensorFeed tools (live AI status, model pricing, GPU pricing, news feeds, routing, more) to any MCP client: Claude Desktop, Claude Code, Cursor, Cline, Continue. Reference example of how an MCP server can wrap an x402-payable HTTP API. Install in one command.

// claude_desktop_config.json { "mcpServers": { "tensorfeed": { "command": "npx", "args": ["tensorfeed-mcp"] } } }
tensorfeed-mcp on GitHub bearer-token + payments docspair with x402

Live: the official MCP registry

TensorFeed snapshots the official registry at registry.modelcontextprotocol.io daily and exposes the rolled-up summary at the public endpoint below. Total servers, daily growth, top namespaces. Pull the JSON for your own agent dashboards.

What an MCP server actually does

Three primitives, JSON-RPC over stdio (local) or SSE (remote). That is the whole protocol.

Tools

Function-call-style endpoints the agent can invoke. Each declares a JSON schema for arguments and return type. Agents pick from this menu when deciding what to call.

Resources

Read-only artifacts the agent can pull. Files, database rows, cached API responses. Surfaced as URIs the agent can dereference.

Prompts

Pre-templated request shapes the user can pick from a menu. The server-side equivalent of a slash-command library.

Who speaks MCP

MCP is the rare standard with first-party support across every major frontier lab and IDE-native agent client. The interop is real, not aspirational.

Anthropic

Origin. Claude Desktop, Claude Code, and the Anthropic SDK speak MCP natively.

OpenAI

Responses API and the Agents SDK ship MCP support across GPT-5.5 and beyond.

Google

Gemini and the Vertex AI agent runtime added MCP in early 2026.

Microsoft / GitHub

Microsoft 365 Copilot, GitHub Copilot, and VS Code agent mode all use MCP for tool integration.

Cursor / Cline / Continue

IDE-native agent clients with built-in MCP server browsers.

Palo Alto Portkey

Enterprise MCP gateway processing trillions of tokens per month, acquired by Palo Alto Networks.

Curated MCP server catalog

The 50-line server pattern

A useful MCP server is roughly a 50-line file. Pick a transport, declare your tools with their JSON schemas, implement the handlers, register the server. The TypeScript skeleton looks like this:

server.ts
import { Server } from '@modelcontextprotocol/sdk/server/index.js';
import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';

const server = new Server(
  { name: 'my-server', version: '1.0.0' },
  { capabilities: { tools: {} } }
);

server.setRequestHandler('tools/list', async () => ({
  tools: [{
    name: 'get_thing',
    description: 'Returns a thing',
    inputSchema: { type: 'object', properties: { id: { type: 'string' } } },
  }],
}));

server.setRequestHandler('tools/call', async (req) => {
  if (req.params.name === 'get_thing') {
    const data = await fetch(`https://api.example.com/thing/${req.params.arguments.id}`);
    return { content: [{ type: 'text', text: await data.text() }] };
  }
  throw new Error('unknown tool');
});

await server.connect(new StdioServerTransport());

Ship that file, publish to npm, list on the registry, your tool is now callable from every MCP client in the ecosystem. Full walkthrough in our 50-line MCP server piece.

Recent MCP coverage on TensorFeed

MCP Just Hit 97 Million Installs. The Agent Era Is Here.

Mar 23, 2026

Anthropic's Model Context Protocol went from experimental to foundational infrastructure. Every major AI provider now ships MCP support.

Palo Alto Just Bought the MCP Gateway. Enterprise Security Has Entered the Agent Stack.

May 1, 2026

Palo Alto Networks acquired Portkey, plugging an AI gateway and an MCP gateway processing trillions of tokens per month into Prisma AIRS.

An MCP Server Is a 50-Line File. Why Every Paid API Should Ship One.

Apr 27, 2026

The actual code, what it costs to ship, and why most teams overthink the work. Stop writing the planning doc; write the file.

MCP + x402: the composable pair

MCP and x402 are complementary, not competitive. MCP is how agents discover and call tools. x402 is how agents pay for them when the tool is gated. The TensorFeed pattern is a worked example: an MCP server (tensorfeed-mcp) wraps an x402-payable HTTP API. The agent calls a tool. The tool consumes a bearer token that was paid for in USDC on Base over x402. Same loop, different layers. MCP solves discovery and shape; x402 solves settlement. The two together are the agent-native API stack.

The x402 hub

FAQ

What is MCP (Model Context Protocol)?

MCP is an open protocol introduced by Anthropic in late 2024 for connecting AI agents to external tools, data sources, and APIs. It defines a JSON-RPC interface (over stdio or SSE) where an agent client speaks to one or more servers, each exposing a structured set of tools (function calls), resources (read-only data), and prompts (templated requests). The same server works in any compatible client: Claude Desktop, Claude Code, Cursor, Cline, Continue, OpenAI Agents SDK, Google Gemini, and more.

How widely is MCP adopted?

As of May 2026, the official registry at registry.modelcontextprotocol.io has 7,985 servers and is adding roughly 270 per day. The Microsoft Build 2025 announcement and broad first-party support across Anthropic, OpenAI, Google, Cursor, Cline, GitHub Copilot, and Microsoft 365 Copilot turned MCP into the default integration layer of the agent stack. Enterprise governance vendors (Palo Alto Portkey acquisition, Datadog, Sentry) entered the layer in early 2026.

What does an MCP server actually do?

An MCP server exposes three primitives. Tools: function-call-style endpoints the agent can invoke (e.g., search_news, get_gpu_prices). Resources: read-only artifacts the agent can pull (e.g., a file, a database row, an API response). Prompts: pre-templated request shapes the user can pick from a menu. The server runs as a child process the client launches, communicating over stdio (local) or SSE (remote). The transport is JSON-RPC 2.0 with a small set of method names defined by the spec.

How long does it take to ship an MCP server?

A working MCP server is roughly a 50-line file. Pick a transport (stdio for local, SSE for remote), declare your tools and their JSON schemas, implement the handler functions, register the server. The TensorFeed reference at github.com/RipperMercs/tensorfeed-mcp wraps 14 paid HTTP endpoints in tools that any MCP client can invoke. The agent-acquisition leverage of having one is far higher than the engineering cost; most teams overthink the work.

How do I add an MCP server to my client?

For Claude Desktop or Claude Code, edit the mcpServers section of your client config (Claude Desktop: claude_desktop_config.json; Claude Code: settings.json) to add the server name and launch command (typically npx @org/server-name). Cursor and Cline have GUI installers that read from the same registry. The official registry at registry.modelcontextprotocol.io has copy-paste install snippets for every published server.

Why should every paid API ship an MCP server?

Agent buyers select tools through MCP. Your paid API is invisible to a Claude Desktop user, a Cline user, or a Cursor user unless you have an MCP server on the registry. The agent decides what to call based on what is registered, not based on the existence of your HTTP endpoint. Shipping an MCP server is the agent-acquisition equivalent of being on Google in 2002 vs. having a website nobody can find. The cost is one afternoon. The reach is every client in the ecosystem.

What is the relationship between MCP and x402?

They compose. MCP is how agents discover and call tools. x402 is how agents pay for them when the tool is gated. The TensorFeed pattern: an MCP server (tensorfeed-mcp) wraps the x402-payable HTTP API (tensorfeed.ai/api/premium). Agent calls the MCP tool. Tool consumes a bearer token paid for via USDC on Base over x402. Same loop, different layers. MCP solves discovery and shape; x402 solves settlement.

For agents: registry snapshot at /api/mcp/registry/snapshot (free, no auth, refreshed daily). Curated server catalog at /api/mcp-servers. Filter with ?capability=... or ?first_party=true.

Further reading