GitHub - aescle/productivity-protocols: An open, evidence-graded bank of health and performance protocols. Queryable by agents over MCP.

Personal productivity protocols, or behaviors you control to get more done. Runs the whole gamut, from peer-reviewed to anecdotal, physical and mental.

You’ll have to experiment to find out what works best for your body. It’s definitely worth doing—it helps in all aspects of life, and you’ll feel a lot better and happier overall.

— Sam Altman

Browse at protocols.aescle.com.

Built by the team behind Aescle.

Install

Claude Code

claude mcp add productivity-protocols -- npx -y productivity-protocols

ChatGPT

Settings -> Connectors -> Add custom connector -> https://protocols.aescle.com/mcp

Example

> my sleep is all over the place on weekends and i feel like shit by monday

⏺ apple-health - health_query (MCP)
  ⎿  14 nights, wake time 06:28 to 07:44

⏺ productivity-protocols - list_protocols (MCP)
  ⎿  227 protocols

⏺ productivity-protocols - get_protocol (MCP)
  ⎿  sleep_regular_wake_time, sleep_precision

⏺ Your wake time is swinging about 75 minutes across the week, which is enough
  to explain the Monday fog on its own. There's a protocol for exactly that.

  45-Minute Wake Window: keep wake time within a 45-minute window.

  Graded moderate. Consistent sleep and wake timing is associated with better
  health, safety, and performance.
  https://pubmed.ncbi.nlm.nih.gov/33054339/

  There's a stricter 20-minute version on the same ladder, but it's only
  graded emerging, so start with the 45.

What's in it

Protocols	227
Evidence grades	17 strong, 52 moderate, 80 emerging, 78 anecdotal
Progression ladders	12 families, 36 laddered protocols
Standalone protocols	191
With contraindications	26
With match triggers	227

A protocol is a behavior, not advice prose. Each one is a single JSON file that a person or an agent can act on.

Evidence grades are honest. Science-backed staples and popular-but-thin founder practices live in one bank without pretending to be the same thing. 78 entries are labeled anecdotal because that is what they are.

The center of gravity is work performance. Longevity levers earn their place by protecting the ability to keep performing, and their benefits are framed that way.

Work kinds: priorities, decisions, starting, focus, rhythm, ideas, learning, review, calendar_defense, social. Body kinds: sleep, training, recovery, nutrition, walk.

Sources span the performance canon: founder essays, the Huberman and Attia literature, the Founders-podcast roster, athletes, and the deep-work research. Every URL is verified live and every PMID is checked against NCBI.

Tools

Tool	Returns
`list_protocols`	The catalog. All 227 by default, about 20k tokens. Optional `kind`, `family`, `grade` filters.
`get_protocol`	Full records for the ids you chose: rationale, contraindications, evidence, citations.

The catalog is also an MCP resource at protocols://catalog, for clients that would rather not spend a tool call on it.

Local client config, if you did not use the one-liner above

{
  "mcpServers": {
    "productivity-protocols": {
      "command": "npx",
      "args": ["-y", "productivity-protocols"]
    }
  }
}

Use it as data

npm install productivity-protocols

import { PROTOCOLS, protocolCore, hasProtocol } from "productivity-protocols";

Or read the JSON directly. data/protocols/<id>.json is the source of truth and has no build step.

{
  "id": "post_meal_walk",
  "name": "Post-Meal Walk",
  "title": "Walk after one meal per day.",
  "benefit": "Steadier afternoon energy and focus.",
  "kind": "walk",
  "family": "meal_timing",
  "tier": "intermediate",
  "evidence": {
    "grade": "strong",
    "summary": "Walking soon after eating reduces acute post-meal glucose excursions.",
    "sources": [{ "label": "Post-meal exercise meta-analysis", "url": "https://pubmed.ncbi.nlm.nih.gov/36715875/" }]
  },
  "indications": ["Focus reliably dips in the hour after lunch."]
}

Evidence grades

Grade	Meaning
`strong`	Consistent trial, meta-analytic, or guideline-level support.
`moderate`	Supportive trials or robust observational evidence, with limitations.
`emerging`	Early, small, or mixed evidence.
`anecdotal`	Practice-driven with a plausible mechanism; little direct trial evidence.

The evidence.summary sentence is written no stronger than the grade supports.

Indications

indications are the triggers: observable, neutral statements an agent can match against a person's real data, like "Wake time swings by hours between weekdays and weekends." They answer when should I try this, which a flat list of protocols never does. Every protocol has them.

Ladders

Protocols with a family sit on a progression ladder: exactly one beginner, one intermediate, and one advanced rung per family, each with a measurable tierTarget. Ladders exist so nobody maxes out. Mastering a rung promotes you to the next one instead of retiring the topic. Standalone protocols omit family entirely.

Layout

data/protocols/<id>.json — one file per protocol. The source of truth.
schema/protocol.schema.json — the contract (JSON Schema draft-07, field docs included).
src/index.js — entry point: PROTOCOLS, protocolCore(id), hasProtocol(id).
src/protocols.generated.js — committed plain-JS mirror, for bundlers and Node.
mcp/handlers.mjs — the MCP tool logic, shared by both transports.
mcp/server.mjs — stdio transport, for local clients. This is what npx runs.
api/mcp.mjs — Streamable HTTP transport, for cloud clients like ChatGPT. Stateless, no auth.
site/index.html — the committed static browse page. Host it anywhere; no build, no backend.
scripts/validate.mjs — zero-dependency validation of the whole bank.

Both transports call the same rpc() in mcp/handlers.mjs, so the tool surface cannot drift between them. Zero dependencies throughout.

Self-hosting

vercel.json deploys site/ as the static root with the MCP endpoint at /mcp. Fork it and you have your own copy of both, with no configuration. Everything is read-only public data, so there is no session state, no auth, and nothing to provision.

Contributing

Contributions are welcome, especially new protocols with real citations and honest grades. See CONTRIBUTING.md. Every rule below is enforced by scripts/validate.mjs in CI, so a contribution that breaks house style fails the check rather than shipping.

Every protocol reads in one voice. Plain words, no metaphors, no cleverness.

Field	Form	Example
`name`	Title Case noun phrase, 2-5 words	`Post-Meal Walk`
`title`	One imperative sentence, ends in a period	`Walk after one meal per day.`
`benefit`	Plain outcome phrase, 14 words or fewer; no second-person address	`Steadier energy after lunch.`
`subtitle`	Two or three word Title Case label	`Meal Protocol`
`rationale`	Plain sentences on why it works, no names	`Walking soon after eating blunts the glucose spike...`
`origin`	One sentence naming who is known for it	`Jeff Bezos replaced slides with narrative memos.`
`indications`	Complete sentences describing an observable state	`Focus reliably dips in the hour after lunch.`

Three rules carry most of the weight. Say what the person gets, not how it sounds. Keep attribution in origin so rationale can stay a plain explanation. Never use a semicolon or a colon to splice two clauses together.

Who maintains this

Built and maintained by the team behind Aescle, health and productivity for top performers. Aescle combines these protocols with your live behavior to help you do your best work every day.

License

Code is MIT (LICENSE). The protocol dataset is CC BY 4.0 (LICENSE-DATA.md).

Not medical advice. Protocols describe general practices and their evidence, not recommendations for any individual. Check avoidWhen where present and talk to a doctor before changing anything that interacts with a medical condition or medication.