Adrian Cole codefromthecrypt

MCP Gateway Documentation Template & Verification Process

Quick Start

Replace [PROJECT_NAME], [PROJECT_URL], and [GITHUB_URL] with the actual project details, then follow the verification process below.

You are a technical documentation auditor specializing in AI infrastructure. Your task is to create a factual, evidence-based documentation section for an MCP (Model Context Protocol) gateway/platform following the strict template and verification process below.

Project to Analyze: [PROJECT_NAME]

WebSocket Transport in MCP: History of Discussions

The Original Decision (PR #206, March 2025)

When @jspahrsummers (Anthropic) proposed replacing the original HTTP+SSE transport with "Streamable HTTP" in [specification PR #206][pr-206], the PR body included an explicit "Why not WebSocket?" section. The core team's stated reasons were:

RPC overhead -- Using MCP in a stateless "RPC-like" way (e.g., a server

title
Rename ACP to IACP

Author(s): @nikomatsakis

Elevator pitch

Rename the Agent Client Protocol from "ACP" to "IACP" (Interoperable Agent Client Protocol) to avoid confusion with OpenAI's [Agentic Commerce Protocol][openai-acp], which launched in September 2025 with backing from Stripe, PayPal, and Salesforce.

ACPs from 2025

These are all I found which include and go beyond https://agentclientprotocol.com/ as of 2025-12-31

SDKs

SDK	Language
[@agentclientprotocol/sdk][ts-npm]	TypeScript
[acpex][acpex-hex]	Elixir

rename elasticsearch to elasticsearch_real and change its port to 9201 rename otel-collector to otel-collector_real and change its port to 4319

add this container with the below flow.py. Then make sure you are sending with javascript as python doesn't support OTEL_EXPORTER_OTLP_PROTOCOL=http/json yet.

then, you can do docker compose logs elasticsearch and see the POSTs.

  elasticsearch:
    image: mitmproxy/mitmproxy

Source diagram of this LI post on goose https://www.linkedin.com/posts/adrianfcole_genai-opensource-activity-7315217199844003840-sX7Q

sequenceDiagram
    participant CLI as Goose CLI
    participant MCP as MCP Server (developer)
    participant LLM

    Note over CLI: User runs `goose run -t "list the files in this directory"`
    Note right of CLI: Forks own binary to start MCP server for "developer" via stdio

	# Add OpenAI and OpenTelemetry ENV variables to .env and run like this:
	# uv run -q --env-file .env main.py
	#
	# Note: Use a larger model like qwen3b:14b, if you are hosting your own models.
	#
	# /// script
	# requires-python = ">=3.13"
	# dependencies = [
	# "openai-agents",
	# "elastic-opentelemetry",

	curl -s -X POST https://api.openai.com/v1/responses \
	-H "Content-Type: application/json" \
	-H "Authorization: Bearer ${OPENAI_API_KEY}" \
	-d '{
	"input": [
	{
	"content": "Which language is this repo written in?",
	"role": "user"
	}
	],

	# e.g. to run with ollama, do this:
	# OPENAI_BASE_URL=http://localhost:11434/v1 OPENAI_API_KEY=unused CHAT_MODEL=qwen3:1.7b uv run a2a.py
	#
	# /// script
	# requires-python = ">=3.13"
	# dependencies = [
	# "a2a-sdk",
	# "uvicorn",
	# "openai",
	# "httpx"

	import os

	import httpx
	from google.adk.agents import Agent
	from google.adk.models.lite_llm import LiteLlm
	from google.adk.runners import Runner
	from google.adk.sessions import InMemorySessionService
	from google.genai import types

	# TODO: native openai https://github.com/google/adk-python/issues/27