Tools · SPBoucher
Menu

Model Context Protocol

In-depth entry

by Model Context Protocol, a Series of LF Projects, LLC (originated by Anthropic) · Specification · also known as MCP, modelcontextprotocol

Open JSON-RPC standard for connecting AI applications to tools, data and prompts through servers, now stewarded under the Linux Foundation with SDKs in many languages

Open sourceOpen sourceFreeLocalSelf-hostedCloudSelf-hostable
Visit official site Compare

Link verified Nov 2, 2026

Overview

The Model Context Protocol (MCP) is an open standard that lets AI applications connect to external systems in a uniform way. Anthropic published it in November 2024; by late 2025 the project had been placed under the Linux Foundation as "Model Context Protocol, a Series of LF Projects, LLC", with a maintainer-led governance model, working groups and a Specification Enhancement Proposal process. New contributions are licensed Apache-2.0 (documentation CC-BY-4.0) while older code remains MIT during a relicensing transition.

The architecture has three roles: hosts (the AI application, such as Claude, ChatGPT, VS Code or Cursor), clients (connectors inside the host) and servers that expose capabilities. Servers offer tools (functions the model may call), resources (data for the user or model) and prompts (templated workflows); clients may offer elicitation, which lets a server ask the user for input. Messages are JSON-RPC 2.0 over stdio for local servers or Streamable HTTP for remote ones, with capability negotiation per session and an authorization specification based on OAuth for remote servers. The current specification revision is dated 2026-07-28 and adds optional extensions: Tasks for long-running asynchronous operations, Skills over MCP for structured agent instructions, and MCP Apps for interactive UI rendered inside conversations. Official SDKs exist for TypeScript, Python and several other languages, and the specification stresses explicit user consent before tool execution.

For a developer, MCP means writing one server for a database, API or internal tool and having it work across every compliant assistant and coding agent. Its limits are those of a young standard: the spec changes yearly, security depends heavily on host implementation, remote-server authorization is still maturing, and tool descriptions are untrusted input that can carry prompt injection.

Editorial note on uncertainty. license_spdx is omitted because the repository is in a documented MIT-to-Apache-2.0 transition with mixed provenance; new contributions are Apache-2.0. The language list reflects SDKs referenced in the official docs and may change.

Capabilities

  • Standard tool, resource and prompt primitives exposed by servers
  • JSON-RPC 2.0 over stdio and Streamable HTTP transports
  • Per-session capability negotiation and OAuth-based authorization for remote servers
  • Elicitation for server-initiated user input
  • Extensions: Tasks (async operations), Skills over MCP and MCP Apps (inline UI)
  • Official SDKs for TypeScript, Python and other languages
  • Adopted by Claude, ChatGPT, VS Code, Cursor and many other hosts
  • Open governance with SEPs and working groups

Typical workflows

  1. Write one MCP server for an internal database and use it from Claude Code, Cursor and ChatGPT
  2. Expose a hosted API as a remote server with OAuth so a team can connect from any client
  3. Package domain instructions as skills discoverable over MCP

Strengths

  • Vendor-neutral foundation governance
  • Broad adoption across assistants and coding agents
  • Simple to implement with the official SDKs
  • Local and remote transports cover desktop and server scenarios

Limitations

  • Specification revised annually; clients and servers may lag
  • Security depends on host-side consent and sandboxing
  • Tool descriptions are untrusted and can carry injection attempts
  • Licensing transition means mixed MIT/Apache-2.0 provenance in older code

Prerequisites

  • An MCP-capable host application
  • A runtime for the chosen SDK (Node.js, Python, etc.)

Who it is for

  • Developers integrating tools and data with AI assistants
  • Vendors exposing their products to agents
  • Platform teams standardizing agent integrations

Alternatives & comparison

MCP standardizes the connection between an agent and its tools, whereas agent frameworks such as the OpenAI Agents SDK or LangChain define the loop that uses those tools; most of them now include MCP client support, so the two are complementary rather than competing. Vendor-specific function-calling schemas remain the alternative when only one host matters. Claude Code is listed here as a reference host that consumes MCP servers.

  • Claude Code

    Anthropic's agentic coding tool that reads a codebase, edits files and runs commands from the terminal, IDEs, a desktop app, the web and CI, with MCP, hooks, skills and subagents

    ProprietaryPaid
  • LangChain

    Open-source Python and JavaScript framework whose create_agent harness builds tool-using agents on the LangGraph runtime with swappable model providers and middleware

    Open sourceOpen source
  • OpenAI Agents SDK

    Lightweight open-source Python and TypeScript framework for agents, handoffs, guardrails, sessions and tracing, defaulting to OpenAI's Responses API with adapters for other providers

    Open sourceOpen source
Compare Model Context Protocol with its alternatives →

Sources & verification

Primary sources consulted for this record. Claims are tied to the fields each source supports; the automated link checker re-verifies every URL on a schedule.

Confidence
High
Reviewed
Nov 1, 2026
Links
Verified · Nov 2, 2026

Spotted an error?

Broken link, outdated license, wrong category, or a tool that no longer exists? Tell us; reports are reviewed by the editor.

Apify

Apify Technologies s.r.o. · Data collection & crawling

In-depth entry

Cloud platform and marketplace for running web-scraping and automation programs (Actors), with storages, proxies, scheduling, an API and an MCP server for agents

Apache-2.0FreemiumUsage-basedNov 2, 2026

Claude Platform

Anthropic · Model providers & APIs

In-depth entry

Anthropic's developer platform for the Claude models: Messages API, tool use, structured outputs, prompt caching, batch processing and Managed Agents, plus cloud availability

ProprietaryUsage-basedEnterpriseNov 2, 2026

LangGraph

LangChain, Inc. · Agent frameworks & orchestration

In-depth entry

Low-level orchestration framework and runtime for stateful, long-running agents expressed as graphs, with durable execution, checkpointing, human-in-the-loop and streaming

MITOpen sourceFreeLocalSelf-hostedNov 2, 2026

OpenAI API

OpenAI · Model providers & APIs

In-depth entry

OpenAI's hosted developer platform: Responses and Chat Completions APIs, Realtime voice, embeddings, images, audio, fine-tuning, evals and managed agents

ProprietaryUsage-basedEnterpriseNov 2, 2026

Pydantic AI

Pydantic Services Inc. · Agent frameworks & orchestration

Type-safe Python agent framework from the Pydantic team with validated structured outputs, dependency injection, swappable model providers, MCP, durable execution and evals

MITOpen sourceFreeLocalSelf-hostedNov 2, 2026

LlamaIndex

LlamaIndex, Inc. · Agent frameworks & orchestration

Open-source Python framework for retrieval-augmented agents over your data, with connectors, indexes, query engines and event-driven workflows, alongside the commercial LlamaParse document platform

MITOpen sourceFreeLocalSelf-hostedNov 2, 2026