Philipp Schmid proposes MCP-based multi-agent architecture
The non-obvious takeaway is that multi-agent orchestration can be treated as an interface problem solved with MCP, not a feature to bake into a harness. Developers building agent frameworks or coding assistants can test the pattern by wrapping their own CLI or loop as an MCP server and using events for subscriptions. Teams evaluating frameworks like Claude Code or custom harnesses gain a simpler upgrade path when coordination logic stays decoupled from the core runtime.
Philipp Schmid argued in a post on X that agent harnesses should drop their built-in multi-agent features and instead expose the harness itself as an MCP server. The core idea: reuse the same loop or CLI recursively to spawn new threads and search within them, with MCP events handling subscriptions between agents.
In this design, different models can take different roles: one model could run a message-board style collaboration, while another follows an orchestrator pattern. Because coordination happens through a standard MCP interface rather than hard-coded multi-agent logic in the harness, teams can iterate on or upgrade individual components without rewriting the harness.
The proposal reflects a broader shift toward MCP as the default protocol for composing agent capabilities, favouring composable, externally addressable primitives over monolithic built-in features.
Agent harnesses should replace built-in multi-agent features with an MCP servers of the harness itself. Use the same loop/cli recursively to create new threads ... or search in them. Use MCP events for subscriptions. You could rebuild message board collaboration for model 1 and have orechstrator pattern for model 2 Much easier to iterate or upgrade.📊 16 ⚡
This report is categorized under AITOP's "How-to" desk, focusing on Europe's AI governance and technology ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include MCP and tools, agents, Phil Schmid, agent harness.
Source attribution: curated from Philipp Schmid (@_philschmid).