One execution surface behind the harness you choose
2ot is the tooling layer underneath scripts and LLM agents. The engineer chooses the harness and workflow; that harness invokes the same routed command model and consumes structured results.
Bring your own harness
2ot deliberately does not prescribe the primary user interface. A harness can be an internal engineering tool, an automation script, a CI job, an agent runtime or a vendor-side integration. Engineers configure and control that workflow, while the harness translates prompts or automation into 2ot operations and presents the resulting diagnostics, changes and build outcomes where the engineer already works.
The stable router stays in front
Automation can use 2ot for workspace operations, fast diagnostics, builds, transpilation and vendor-specific routes. A configured workspace lets the router infer the vendor for common commands. Direct terminal use remains useful for setup, inspection and debugging, but the same command surface is designed to sit behind higher-level harnesses.
The same surface becomes structured tools
Schema-emitting 2ot commands can be discovered dynamically and normalized into MCP tool definitions. That lets an LLM harness call the execution surface directly instead of depending on a second, hand-maintained automation API.
Dynamic discovery instead of duplicated APIs
The router asks compatible leaf executables for their schemas, normalizes command names and arguments, and emits one machine-readable tool surface.
The MCP runtime discovers that schema and exposes the resulting tools to compatible AI harnesses while preserving the same routing boundaries as the CLI.
Fast diagnostics for agent loops
2ot check --fast gives a harness a low-latency compiler feedback loop without starting vendor software. Vendor-aligned diagnostic mappings and structured enrichments make the result more useful for explanation and repair than a bare compiler message.
Compiler context for repair
2ot-agent-context packages compiler and semantic context around source and diagnostics. It complements the routed CLI tools so an LLM can reason about the problem with more than an isolated error string.
Discovery starts before an agent connects to the local toolchain
The public site now exposes a separate read-only discovery layer: /llms.txt,/llms-full.txt, a crawler sitemap and WebMCP tools for documentation and capability lookup. Those surfaces advertise what 2ot can do without turning the marketing site into a remote industrial control plane. Operational MCP remains local to the harness and environment chosen by the engineer.
Vendor-specific operations still stay vendor-specific
MCP does not flatten Siemens TIA, CODESYS and TwinCAT into a fictional lowest-common denominator. Common workspace commands can use configured routing, while explicit vendor routes remain available when the requested operation depends on a particular engineering environment.