LIVE
ANTHROPICOpus 4.7 benchmarks published2m ago
CLAUDEOK142ms
OPUS 4.7$15 / $75per Mtok
CHATGPTOK89ms
HACKERNEWSWhy has not AI improved design quality the way it improved dev speed?14m ago
MMLU-PROleader Opus 4.788.4
GEMINIDEGRADED312ms
MISTRALMistral Medium 3 released6m ago
GPT-4o$5 / $15per Mtok
ARXIVCompositional reasoning in LRMs22m ago
BEDROCKOK178ms
GEMINI 2.5$3.50 / $10.50per Mtok
THE VERGEFrontier Model Forum expansion announced38m ago
SWE-BENCHleader Claude Opus 4.772.1%
MISTRALOK104ms
ANTHROPICOpus 4.7 benchmarks published2m ago
CLAUDEOK142ms
OPUS 4.7$15 / $75per Mtok
CHATGPTOK89ms
HACKERNEWSWhy has not AI improved design quality the way it improved dev speed?14m ago
MMLU-PROleader Opus 4.788.4
GEMINIDEGRADED312ms
MISTRALMistral Medium 3 released6m ago
GPT-4o$5 / $15per Mtok
ARXIVCompositional reasoning in LRMs22m ago
BEDROCKOK178ms
GEMINI 2.5$3.50 / $10.50per Mtok
THE VERGEFrontier Model Forum expansion announced38m ago
SWE-BENCHleader Claude Opus 4.772.1%
MISTRALOK104ms
Back to AI agents

MCP (Model Context Protocol)

The open standard for exposing tools, resources, and prompts to AI agents. Built by Anthropic and adopted by Claude Desktop, Claude Code, and a growing ecosystem of agentic applications.

Origin

MCP was introduced by Anthropic in late 2024 to solve a recurring problem: every LLM application was reinventing tool exposure. Plugins for OpenAI, Functions for Anthropic, custom adapters for everything else. MCP defines one protocol so an LLM-powered app like Claude Desktop can speak to any compliant server, and any developer can write one server that works across compatible clients.

The model

An MCP server exposes three primitives:

  • Tools: callable functions the LLM can invoke (with JSON Schema inputs)
  • Resources: structured content the LLM can read (files, API payloads, DB rows)
  • Prompts: reusable templates the user can invoke

The client (Claude Desktop, Claude Code, etc.) connects to the server via stdio, SSE, or streamable HTTP. The model reads the tool descriptions, picks ones to call, and the server returns structured content the model uses in its reply.

Discovery

The official MCP registry at registry.modelcontextprotocol.io is a centralized directory of community servers. Servers declare metadata via a server.json manifest and publish via the mcp-publisher CLI. There's also a robust ecosystem of curated lists (punkpeye/awesome-mcp-servers, wong2/awesome-mcp-servers, and many others on GitHub).

TensorFeed's MCP server

TensorFeed publishes @tensorfeed/mcp-server with 20 tools (5 free, 15 premium). Free tools cover real-time AI news, service status, model pricing, benchmarks, and a today summary. Premium tools cover routing recommendations, news search, history series, cost projection, forecasting, provider deep-dive, model comparison, the agent morning brief, and webhook watch management.

{
  "mcpServers": {
    "tensorfeed": {
      "command": "npx",
      "args": ["-y", "@tensorfeed/mcp-server"],
      "env": { "TENSORFEED_TOKEN": "tf_live_..." }
    }
  }
}

Without the token, only free tools work. With the token, premium tools unlock. The server is registered in the official MCP registry as ai.tensorfeed/mcp-server.

Why MCP matters for agent payments

MCP standardizes how an agent discovers and calls tools. When you combine that with x402 (which standardizes how a tool gets paid), you get a complete machine-to-machine commerce layer. The agent finds a tool via MCP, the tool returns 402, the agent pays on-chain, the tool serves the data. No human in any step.

Further reading