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.

Carnitas braise "until they shred", which nobody can put a number on, and the salsa, the margaritas, the tortillas and the table all have to land with them. One blender, two burners. Here is that dinner as a live timeline:

python3 -m venv rhylthyme-env && source rhylthyme-env/bin/activate   # Python 3.12 or newer
pip install rhylthyme
rhylthyme publish https://raw.githubusercontent.com/rhylthyme/.github/main/profile/examples/taco-night.json --image taco-night.png --open

(Or install nothing: put uvx --python 3.12 --from rhylthyme-cli-runner in front of it.) No account, no API key. It prints a summary, a text chart and

Live timeline: https://kitchen.rhylthyme.com?share=c83cdcd3da455fc8
Picture: taco-night.png

Open the link on your phone: an ingredient checklist that doubles as a shopping list, an itinerary with the method under each step, and a timeline with timers and audio cues. When the pork finally shreds you end the braise and everything after it moves. The picture:

The hatched bar is the braise, which ends when you say so. The dashed arrows are steps that start a set time before it is due to end (char the tomatoes 30 minutes out, juice the limes 12 minutes out), so if the pork needs another quarter of an hour, the margaritas wait with it. Now make it yours: save the program, add guacamole or take away a burner, and rhylthyme publish taco-night.json. Or start from a recipe you did not write: rhylthyme import https://www.seriouseats.com/the-best-chili-recipe --publish turns a page from any of about 580 recipe sites into a timeline (TheMealDB, Spoonacular, CookLang, protocols.io, Opentrons and Benchling work the same way). Other ways to make one: working back from a deadline in a terminal, rhylthyme-cli-runner; by asking Claude or ChatGPT, rhylthyme-mcp.

Start from something already written: a recipe page, a protocols.io protocol, an Opentrons .py file, a CookLang file, a Benchling protocol.

pip install rhylthyme
rhylthyme import https://www.bbcgoodfood.com/recipes/classic-lasagne --publish   # any of ~580 recipe sites
rhylthyme import 52772 -i themealdb                                               # a TheMealDB id -> teriyaki_chicken_casserole.json
rhylthyme import https://raw.githubusercontent.com/Opentrons/Protocols/develop/protocols/007992/rna_isolation.ot2.apiv2.py   # an Opentrons protocol
rhylthyme importers                                                               # what is installed

import validates the program, writes <programId>.json, and with --publish prints a live-timeline URL. rhylthyme render <file>.json -o fig.png draws it. Details: rhylthyme-importers.

In Claude Code:

/plugin marketplace add rhylthyme/rhylthyme-mcp
/plugin install rhylthyme@rhylthyme

then ask: "Plan Thanksgiving for 8 with one oven, eating at 6 pm." In Claude, ChatGPT or Cursor, add https://mcp.rhylthyme.com/mcp as a connector instead. No account is needed; the answer is a live-timeline link.

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 Code plugin. One install connects the hosted server and adds a skill that teaches Claude to write a schedule well (extract the steps before relating them, validate, check for conflicts, work back from a deadline):

/plugin marketplace add rhylthyme/rhylthyme-mcp
/plugin install rhylthyme@rhylthyme

Any other MCP client. Add an endpoint URL as a connector: Claude (Settings → Connectors → Add custom connector), ChatGPT (developer mode → create a connector), Cursor ({"url": "https://mcp.rhylthyme.com/mcp"}), or

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

Clients that can only launch a command: pip install rhylthyme-mcp gives a rhylthyme-mcp stdio bridge to the hosted server. No MCP client at all: a single JSON-RPC POST works with no handshake, and llms.txt on every rhylthyme.com host says how.

Streamable HTTP, stateless. No account or API key is needed for validation, analysis, publishing a timeline or the public catalog; a personal library and recorded runs use your Rhylthyme account through OAuth 2.1. See rhylthyme-mcp for the tool reference.

pip install rhylthyme installs rhylthyme-cli-runner, rhylthyme-importers and rhylthyme-timeline together. rhylthyme-mcp and rhylthyme-cli-runner are easy to confuse: the first is what an assistant calls, the second is what you run yourself on a program file, and the second is one of the first's clients. Each README has a side-by-side table.

The web application lives in rhylthyme-server, which is being prepared for public release.

pip install rhylthyme          # the rhylthyme command, the importers and the renderer
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 throughout: the specification, tools, examples, timeline engine and MCP server.