The cookbook for the Stanley MCP: run your content operation from a conversation.

Quickstart · How it works · Recipes · Skills · Site · Contributing · Changelog

What is this?

stanley-mcp-playground is a cookbook for Stanley, an AI content-marketing agent, connected to Claude over MCP. Stanley handles voice, calendar, publishing, and analytics; Claude handles orchestration: reading your commits and meeting notes, pulling assets from design tools, chaining steps, running things on a schedule. Every recipe here lives in the seams between the two, and every prompt in them has been run against a real account before landing.

The cookbook teaches nine workflows, in learning order: connect and audit, plan a week in one sitting, turn merged PRs into build-in-public posts, review your own numbers, scout trends, repurpose across platforms, attach media from any outside tool, and put all of it on a schedule.

Quickstart

git clone https://github.com/JayantChopra/stanley-mcp-playground.git
cd stanley-mcp-playground
claude

Approve the project MCP server when Claude Code asks, sign in with /mcp, then say "where do I start". Claude audits what you've connected and points you at the recipes your setup can run.

On claude.ai instead: add https://xapi.getstanley.ai/mcp as a custom connector (Settings → Connectors) and paste prompts straight from the recipes. You'll need a Stanley account with at least one platform linked, from getstanley.ai.

How it works

 you ──▸ Claude ──▸ message_stanley ──▸ Stanley ──▸ X · LinkedIn · Threads · Instagram · Substack Notes
           │                              │
      your other tools               calendar · analytics · media · rituals
      (GitHub, Canva, Notion…)

On Claude's side, Stanley is a single MCP tool with persistent conversation state, so recipes are conversation patterns rather than API call sequences. The full capability sheet, limits, and approval model live in recipe 09.

Recipes

#	Recipe	In one line
01	Connect and run your first loop	MCP setup, connection audit, draft then schedule
02	Plan a week in one sitting	fill your calendar slots from one working session
03	Build in public	commits and PRs turned into story posts
04	Analytics review	SQL over your own post metrics, fed back into planning
05	Trend scouting	niche trends, video teardowns, creator audits
06	Repurposing	one piece of content, every platform
07	External media	Canva, Recraft, or anything with a URL, attached to drafts
08	Rituals	morning briefs, cron jobs, poll-based triggers
09	What Stanley can do	the reference sheet the other eight lean on

The same recipes render at stanley-cookbook.vercel.app, with copy buttons on every paste-able prompt.

Skills

.claude/skills/ ships five slash commands that work the moment you open this repo in Claude Code:

Skill	What it does
`/stanley-start`	audits your connections, recommends the recipes your setup can run
`/stanley-week-plan`	plans and bulk-schedules a week of posts interactively
`/build-in-public`	scans your GitHub activity and drafts posts from what you pick
`/stanley-review`	runs the weekly analytics review as a short digest
`/repurpose`	converts one post into drafts for every platform you name

Copy any folder into ~/.claude/skills/ to use it outside this repo. Every skill carries the same hard rules: nothing publishes live, gets deleted, or goes out as a DM or email without your explicit approval; scheduling drafts is allowed because it's reversible.

Repository layout

recipes/          the nine recipes, numbered in learning order
.claude/skills/   drop-in slash commands
site/             the static site, generated from recipes/ by tools/build-site.mjs
tools/            prose lint + site builder
.mcp.json         registers the Stanley MCP for anyone who clones
CLAUDE.md         tells Claude to audit connections and route, not guess

Status

Recipes 01 through 09 are complete and tested against a live account. Planned next: worked examples with real (redacted) output in every recipe, a meeting-notes-to-post recipe, and a monitoring digest recipe. Track CHANGELOG.md.

Community

Bugs and recipe proposals: GitHub Issues, templates provided
Security: see SECURITY.md; in this repo, wording that weakens the approval model counts
Conduct: CODE_OF_CONDUCT.md

Contributing

Contributions are welcome, with one non-negotiable: run your recipe against a real Stanley account before you submit it. Start with CONTRIBUTING.md; CI enforces the writing standard, so run python3 tools/lint.py before pushing.

License

Apache License 2.0.