Model Context Protocol (MCP) is the rare moment of cross-vendor convergence. Anthropic introduced it in late 2024; OpenAI's Codex, Google's Gemini, and dozens of third-party tools adopted it through 2025–2026. Today, an MCP server you write once works in Claude Code, Cowork, Codex, Gemini, and any other compliant client.
The architecture: servers expose tools (functions with structured I/O), clients (the agents) call them like any tool, transports (stdio for local, HTTP for remote) keep the spec light. There are 1,800+ public servers — Sentry, GitHub, Notion, Slack, Linear, Stripe, Airtable, Postgres, Cloudflare, every major SaaS.
The portability dividend: configure once, use everywhere. The Sentry MCP you wired into Claude Code last month works in Codex tomorrow. Your team's internal MCP server does double-duty across whichever CLI each developer prefers.
Code
Same MCP, three clients·bash
# Claude Code
claude mcp add --transport http sentry https://mcp.sentry.dev/mcp
# Codex
codex mcp add --transport http sentry https://mcp.sentry.dev/mcp
# Gemini
gemini mcp add --transport http sentry https://mcp.sentry.dev/mcp
# In every client:
> "what's the top error in production over the last 24h?"
# Each agent calls Sentry MCP; each gets the same answer shape.
Custom MCP — write once, works everywhere·json
// my-internal-mcp/server.json — minimal MCP server stub
{
"name": "internal-billing",
"version": "1.0.0",
"tools": [
{
"name": "lookup_customer",
"description": "Fetch customer record by id or email",
"inputSchema": { "type": "object",
"properties": { "query": { "type": "string" } } }
},
{
"name": "list_recent_charges",
"description": "List charges in the last N days",
"inputSchema": { "type": "object",
"properties": { "days": { "type": "integer", "default": 7 } } }
}
]
}
Pick one MCP server you use in your primary CLI. Add it to a different CLI you also have installed. Run the same prompt against both; verify the answer matches. The portability dividend just paid out.
Progress
Progress is local-only — sign in to sync across devices.