MCP

The USB-C of agent tooling: a single JSON-RPC interface for connecting any agent to any tool, data source, or service.

name
MCP
full_name
Model Context Protocol
layer
capability
creator
Anthropic (now governed under the Linux Foundation's Agentic AI Foundation)
status
de facto standard
year
2024
one_liner
The USB-C of agent tooling: a single JSON-RPC interface for connecting any agent to any tool, data source, or service.
spec_url
https://modelcontextprotocol.io
snippet
{ "jsonrpc": "2.0", "method": "tools/call", "params": { "name": "search", "arguments": { "q": "..." } } }
abbreviation
MCP
also_known_as
Model Context Protocol
canonical_spec_url
https://modelcontextprotocol.io/specification/2025-11-25
entity_uri
https://en.wikipedia.org/wiki/Model_Context_Protocol
taxonomy_layer
capability
sub_layer
tool-calling
protocol_type
tooling
central_problem
Gives an AI agent a single, uniform interface to call any external tool, data source, or service instead of bespoke per-integration glue.
maintainer
Model Context Protocol project (Anthropic + community), hosted by the Agentic AI Infrastructure Foundation (AAIF) under the Linux Foundation
governance_body
AAIF
license
MIT (spec and SDKs)
maturity_tag
production-ready
current_spec_version
2025-11-25
spec_date
2025-11-25
launch_date
2024-11-25
last_verified
2026-06-15
transport
JSON-RPC 2.0 over stdio or Streamable HTTP
core_mechanism
An MCP client (the agent host) connects to one or more MCP servers; servers advertise tools, resources, and prompts, and the client invokes them with typed JSON-RPC calls (tools/list, tools/call). A November 2025 revision adds an experimental async Tasks primitive.
discovery_endpoint
JSON-RPC initialize + tools/list (server-advertised; no fixed URL path)
settlement_type
adoption_metric
Hosted under AAIF as a load-bearing project; broad SDK and vendor support across the agent ecosystem verify-against-primary-at-build ↗ https://github.com/modelcontextprotocol/modelcontextprotocol
notable_adopters
{"value":"Anthropic","source":"https://modelcontextprotocol.io"}
relationships
{"predicate":"governed_by","target":"aaif","note":"MCP -governed_by-> AAIF (research §2 seed triple)"} {"predicate":"complements","target":"a2a","note":"A2A complements MCP; MCP handles agent→tool, A2A handles agent→agent"} {"predicate":"discovered_by","target":"agents-json","note":"agents.txt/agents.json-class discovery files can point agents to MCP endpoints"}
ideal_use_case
Connecting an LLM agent to your data sources, internal tools, or SaaS APIs through one standard interface.
when_to_use
When agents need to take actions or read structured data from tools/services and you want one integration surface instead of N bespoke ones.
when_not_to_use
When you only need an agent to read content (use Layer-1 discovery files like llms.txt), or when the interaction is agent-to-agent coordination (use A2A).
code_example
{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "search", "arguments": { "q": "agentic web" } } }
source
Spec version/date: https://modelcontextprotocol.io/specification/2025-11-25 (latest stable, 2025-11-25). Governance (AAIF hosts MCP): competitive-research-2026-06.md §2.
agent_readiness_link
agent-readiness/mcp
layer_legacy
tool

← all The Agent Protocol Atlas · .md · JSON