The Model Context Protocol, the standard that lets AI agents reach into tools like Gmail, Slack, and Salesforce, is getting a stateless-session update next week that should make it materially easier to run MCP servers at scale. The official spec has been public since May, but the practical implications were laid out Monday morning by Arcade, a two-year-old startup that raised $60 million in June on the bet that AI agent infrastructure — not model quality — is the bottleneck holding back real deployments.
MCP is the connective tissue for agentic AI. Instead of engineers hand-building a pipe for every integration, a model like Claude speaks MCP to a server, and the server exposes whatever tools and data it wants the model to touch. The protocol has become the default way frontier labs, tool vendors, and enterprises wire agents to real systems, which is why a change to how it handles sessions matters.
Under the current design, an MCP server hands the client a session ID on first contact and expects to see that same ID on every subsequent request. That works fine for a demo. It works badly for a company trying to serve millions of users through a load balancer that routes requests across dozens of servers in different regions.
Key facts
- 01The Model Context Protocol update ships next week, moving session handling from stateful IDs to a stateless design.
- 02The official spec has been public since May, but adoption details were laid out Monday by Arcade founder Nate Barbettini.
- 03Arcade raised $60 million in June on the thesis that most AI agents fail on infrastructure, not model quality.
- 04MCP is the plumbing that lets clients like Claude reach into Gmail, Slack, and Salesforce without custom-built pipes for each service.
- 05The stateless approach mirrors how ordinary websites already work behind load balancers, cutting server-side coordination overhead.
Arcade founder Nate Barbettini walked through the mechanics in a public explanation Monday. The client sends a hello, the server replies with capabilities and a session ID, and every request afterward has to carry that ID back to the specific server that issued it. If the ID expires, the client has to notice and re-handshake.
The problem is that stateful sessions assume one server remembers you. Modern deployments do not work that way. Traffic gets sprayed across a farm of interchangeable machines, and any given request might land on a server that has never seen the caller before. Making that work today means bolting on shared session stores, sticky routing, or region-pinning — engineering work that fights the load balancer instead of using it.
That has been a real drag on first-party MCP adoption. Despite the volume of hype around agentic AI this year, relatively few large companies have shipped their own MCP servers at scale. The session-ID plumbing is a big part of why.
The new spec takes a stateless approach on the server side, closer to how ordinary web APIs already operate. Servers stop being responsible for remembering who's who across requests, which means any machine in the farm can handle any request without coordinating with its neighbors. Cheaper to run, easier to scale, and friendly to the load balancers companies already have deployed.
Arcade has a direct commercial interest in this working. Its product line depends on agents reliably connecting to enterprise tools like Gmail, Slack, and Salesforce, and every session-management hack the company has to write for customers is a cost it would rather not carry. A cleaner protocol lifts Arcade and every other MCP-native vendor at the same time.
The counterweight is that stateless designs push some complexity back onto the client and onto whatever the client uses to persist context. Sessions do not disappear — they move. Real-world MCP deployments will still need to solve auth, rate limiting, and continuity across long-running agent runs, and doing that on the client side has its own failure modes. The update simplifies the server operator's life; it does not eliminate the underlying problem of maintaining state across a multi-turn agent conversation.
It is also a reminder that protocol work moves at a different clock speed than model releases. Frontier labs ship new capabilities every few weeks. Standards bodies rewrite session handling on a timeline measured in quarters. The infrastructure that agents depend on is real engineering, and it advances through log-rolling consensus rather than benchmark leaderboards.
For the AI market, the read-through is straightforward: the value in agentic AI is shifting toward whoever solves the boring parts. Model quality is now good enough that the constraint on shipping real agent products is integration reliability, session state, auth flows, and the operational cost of running MCP servers for millions of users. A stateless MCP makes that stack cheaper to operate, which is why the companies that stand to benefit most are the ones — like Arcade — whose business model already assumes the models are fine and the plumbing is the moat.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
