Stateless MCP has recaptured my interest (and inspired mcp-explorer and datasette-mcp)

Simon Willison · Simon Willison · 2026-07-31

Simon Willison argues that the MCP 2026-07-28 stateless specification simplifies agent tool protocols from two HTTP requests to one, and announces three new tools—mcp-explorer, datasette-mcp, and llm-mcp-client—that demonstrate the approach is safer than giving agents arbitrary shell access.

Open original ↗

Appears in

MCP Protocol Redesigned as Stateless to Unlock Enterprise Adoption

Extraction

Topics: model-context-protocolagent-securityagent-toolingdatasettellm-tooling

Claims

Stateless MCP (MCP 2026-07-28) reduces tool invocation from two HTTP requests requiring session management to a single self-contained request.
MCP tools are easier to audit and control than granting agents arbitrary shell and internet access, and are small enough for laptop-class models to drive.
The stateless spec makes MCP better suited for scalable web applications because servers no longer need to maintain per-session state or route sticky sessions.
datasette-mcp exposes list_databases, get_database_schema, and execute_sql as MCP tools on any Datasette instance.
MCP provides a fundamentally safer foundation than arbitrary command execution for building sensitive LLM-powered applications.

Key quotes

MCP tools are easier to audit and control, and simple enough that smaller models that run on a laptop can still drive them reasonably well.

Something I've come to appreciate about MCP is that it's much easier to reason about agent capabilities and what might go wrong than with arbitrary command execution in an open network environment - the default for most of today's general and coding agent tools.

I plan to lean into MCP a whole lot more when I'm building sensitive applications on top of LLMs.