MCP Dev Summit opens in Toronto with coordination still unsolved

MCP Dev Summit opens today, Monday, October 5, in Toronto. The news comes from a Forkast report picked up by Google News, with a headline that sets the tone for the event: the agent protocol stack is looking for its missing coordination layer.

It is worth being clear about what we have. At the time of writing we have only been able to read that headline, not the full text, and we do not know the agenda or the speaker list. We are not going to fill those gaps with guesses. What we can do is explain what the idea of a missing layer refers to, because it is a discussion that people working with MCP have been having for some time.

Which stack we are talking about

MCP was born in November 2024, when Anthropic released it as an open standard so that a model could call external tools and data sources through a common interface. In late 2025 the protocol moved to the Agentic AI Foundation, under the Linux Foundation umbrella, so its governance no longer depends on a single company.

MCP solves one specific relationship: that of a model with its tools. For the relationship between agents there is A2A, the protocol Google introduced in 2025 and which also ended up at the Linux Foundation. With those two pieces, an agent knows how to use a tool and how to talk to another agent.

What neither of them defines is who hands out the work. When several agents collaborate on a long task, someone has to decide which one handles each part, how they share state, what happens if two of them try to modify the same thing, and how you later reconstruct who did what. That is, broadly, what is meant by coordination.

How it gets solved today

It gets solved inside each product. In Claude Code, for example, a main agent delegates to specialized subagents, skills supply instructions on demand, and hooks let you run commands on events such as PreToolUse or Stop to enforce your own rules. It works well as long as everything happens in the same environment and under the same team.

The problem shows up when you cross boundaries. If one company's agent has to split a task with a vendor's agent, built on a different framework, there is no shared agreement on how to assign work or how to keep a record of it. Every integration is negotiated by hand. It is the same situation tools were in before MCP, one step further up.

The lens to read this news through

Forkast is an outlet specialized in blockchain and digital assets. That is not a minor detail: from that sector, "coordination" often also means identity, a shared ledger and payments between agents. This is our inference and not something we can attribute to the article, but it helps place the headline. The MCP community and the crypto community are not necessarily thinking about the same layer.

Who it matters to and what to watch

For anyone maintaining MCP servers, nothing changes this week: a coordination layer, if it arrives, would sit on top and should not force a rewrite. For anyone orchestrating several agents in production, however, what comes out of Toronto could shape architecture decisions. We will be watching three things:

1. Whether a specification proposal with actual text and use cases is presented, or whether the topic stays at the panel discussion level.
2. Whether the proposal comes through the MCP process or as a separate protocol with its own governance.
3. How it handles authorization and traceability, because coordinating agents is mostly about deciding who can ask what of whom.

Our take

We are in no hurry for one more layer. MCP spread because it solved a bounded problem with a small specification, and a coordination layer will only deserve adoption if it keeps that restraint. Until there are concrete proposals on the table, we still prefer simple, explicit orchestration to a premature abstraction.