Rhylthyme ("real time") is a declarative language and runtime for schedules that a person executes: several lines of work running in parallel, dependencies between them, steps whose length is fixed, bounded or open-ended, and a few shared resources that only so many steps can use at once. A program is JSON; the runtime plays it live with a clock, cues and manual gates; an MCP server lets an AI agent author, validate and publish one as tool calls.

The same schema serves four verticals:

General entry point: rhylthyme.com. Documentation: docs.rhylthyme.com.

A remote Model Context Protocol server exposes validation, timing analysis, catalog search, import and publication as annotated tools, plus the schema, an authoring guide and example programs as resources.

https://mcp.rhylthyme.com/mcp            generic
https://mcp.rhylthyme.com/kitchen/mcp    + cook_recipe, whats_for_dinner
https://mcp.rhylthyme.com/lab/mcp        + run_protocol, random_protocol, Benchling import
https://mcp.rhylthyme.com/events/mcp     + plan_event, random_event_template
https://mcp.rhylthyme.com/gym/mcp        + start_workout, surprise_workout

claude mcp add --transport http rhylthyme https://mcp.rhylthyme.com/kitchen/mcp

Streamable HTTP, no sign-in needed for the public catalog and the pure tools (validate_program, analyze_schedule). See rhylthyme-mcp for the tool reference.

The web application, importers (Spoonacular, TheMealDB, protocols.io, Cooklang, Opentrons, Benchling) and the MCP server implementation live in rhylthyme-server, which is being prepared for public release. A technical report describing the language, runtime and MCP server, and positioning them against project scheduling, temporal reasoning, robotics, lab automation and workflow engines, is in preparation.

pip install rhylthyme-cli-runner rhylthyme-spec
git clone https://github.com/rhylthyme/rhylthyme-examples
rhylthyme validate rhylthyme-examples/programs/breakfast_schedule.json
rhylthyme run      rhylthyme-examples/programs/breakfast_schedule.json

A minimal program:

{
  "schemaVersion": "0.1.0",
  "programId": "eggs-and-toast",
  "name": "Eggs and toast",
  "tracks": [
    { "trackId": "eggs", "name": "Eggs", "steps": [
      { "stepId": "whisk", "name": "Whisk", "task": "prep",
        "duration": { "type": "fixed", "seconds": 60 },
        "startTrigger": { "type": "programStart" } },
      { "stepId": "cook", "name": "Cook", "task": "stove",
        "duration": { "type": "variable", "minSeconds": 120, "maxSeconds": 240, "defaultSeconds": 180 },
        "startTrigger": { "type": "afterStep", "stepId": "whisk" } } ] },
    { "trackId": "toast", "name": "Toast", "steps": [
      { "stepId": "toast", "name": "Toast", "task": "toaster",
        "duration": { "type": "fixed", "seconds": 180 },
        "startTrigger": { "type": "afterStep", "stepId": "cook", "event": "start", "offsetSeconds": 60 } } ] }
  ],
  "resourceConstraints": [
    { "task": "prep", "maxConcurrent": 1 },
    { "task": "stove", "maxConcurrent": 2 },
    { "task": "toaster", "maxConcurrent": 1 }
  ]
}

Steps in a track run one after another; parallel work goes in separate tracks; every task needs a resource constraint; durations and offsets take seconds or strings like "5m".

Apache-2.0 for the specification, tools, examples and timeline engine; the MCP documentation repository is MIT.