plsreadme

The Problem

You wrote a README, a PRD, meeting notes, or an API doc in markdown. Now you need to share it with someone who doesn't have a markdown renderer, doesn't use GitHub, or just needs a clean link they can open in a browser.

plsreadme turns any markdown into a permanent, beautifully rendered web page in one step. No accounts. No sign-ups. No friction.

✨ Features

• Instant sharing — Paste markdown or upload a file, get a plsrd.me link

• Beautiful rendering — Clean typography, dark mode, mobile-responsive

• Inline comments — Readers can click any paragraph and leave feedback

• Review mode (current vs timeline) — Multi-version docs default to Current draft feedback with one-click access to full Timeline history

• AI auto-formatting — Throw raw text at it; it comes out as clean markdown

• MCP server — Share docs directly from Claude, Cursor, VS Code, or any MCP client

• OpenClaw skill — Available on ClawHub for AI agent workflows

• Short links — Every doc gets a compact plsrd.me/v/xxx URL

• Raw access — Download the original .md file from any shared link

• Version timeline + safe restore — /v/:id/versions + /v/:id/history + archive-first restore API for fast rollback

• Clerk auth foundation — GitHub/Google sign-in wiring + Clerk-hosted email fallback + backend auth verification utilities

• Ownership model (Phase 2) — docs can be linked to a Clerk user (owner_user_id) while preserving anonymous flows

• My Links dashboard (Phase 3) — authenticated /my-links page with search/sort/pagination and quick copy/open actions

• Legacy link claiming (Phase 4) — signed-in users can claim older anonymous links by proving the original admin_token

• Zero config website demo — No account or API key needed to try it in the browser

🚀 Quick Start

Web

Go to plsreadme.com, paste your markdown, click share.

Auth Paths And Rollout State

Recommendation order:

1. Try in browser first — fastest demo path, no MCP setup required.

2. Use hosted remote MCP with browser login when client support is verified.

3. Use API key / local MCP fallback when interactive login is unavailable.

Current rollout state:

Hosted remote MCP rollout notes:

• https://plsreadme.com/mcp

• https://plsreadme.com/sse

Those hosted remote MCP routes are live behind OAuth-protected browser login in code, including /authorize, /oauth/token, and /oauth/register.

Operational notes:

• D1 doc_create_events is the canonical create-attribution table across web, hosted MCP, and local MCP flows.

• docs.raw_view_count tracks every render hit, while docs.view_count is reserved for likely-human reads.

• See docs/runbooks/auth-surface-monitoring.md for the production query set and response steps.

• access tokens last about 1 hour

• refresh tokens last about 30 days

• reconnecting the same client replaces the older grant

• signing out of the website does not revoke an existing editor grant by itself

• this repo is now wired to a dedicated Cloudflare Workers KV binding named OAUTH_KV

When browser login is not available in your client, create a personal API key from /my-links and use either the hosted remote header fallback or the local npx -y plsreadme-mcp package.

Website demo trust model today:

• anonymous website creates on /api/create-link require a short-lived browser verification grant

• signed-in website creates skip that grant and stay friction-light

• post-create UI now branches into Save to my account, Connect your editor, and Copy link

API

curl -X POST https://plsreadme.com/api/render \
  -H "Content-Type: application/json" \
  -d '{"markdown": "# Hello World\n\nThis is my doc."}'

{
  "id": "abc123def456",
  "url": "https://plsreadme.com/v/abc123def456",
  "raw_url": "https://plsreadme.com/v/abc123def456/raw",
  "admin_token": "sk_..."
}

Save the admin_token — you'll need it to edit or delete:

# Update
curl -X PUT https://plsreadme.com/v/abc123def456 \
  -H "Authorization: Bearer sk_..." \
  -H "Content-Type: application/json" \
  -d '{"markdown": "# Updated content"}'

# Delete
curl -X DELETE https://plsreadme.com/v/abc123def456 \
  -H "Authorization: Bearer sk_..."

Version timeline + safe restore

Use the timeline endpoint to review revision context during AI iteration cycles:

curl https://plsreadme.com/v/abc123def456/versions

{
  "id": "abc123def456",
  "current_version": 5,
  "total_versions": 5,
  "versions": [
    { "version": 5, "is_current": true, "raw_url": "https://plsreadme.com/v/abc123def456/raw" },
    { "version": 4, "is_current": false, "raw_url": "https://plsreadme.com/v/abc123def456/raw?version=4" }
  ]
}

If an AI edit regresses the doc, restore a prior snapshot (archive-first, non-destructive):

curl -X POST https://plsreadme.com/v/abc123def456/restore \
  -H "Authorization: Bearer sk_..." \
  -H "Content-Type: application/json" \
  -d '{"version": 4}'

Restore is rate-limited similarly to updates (currently 60/hour per actor key) to reduce abuse.

For docs owned by an authenticated Clerk user, update/delete/restore also require that owner session (to prevent cross-user mutation), while anonymous docs continue to work with admin_token only.

Review mode usage notes (Current draft first, Timeline on demand)

The document viewer now exposes comment review controls:

• Current draft — shows only comments tied to the latest doc version (default when a doc has multiple versions).

• Timeline — shows the full cross-version comment history.

You can fetch the same modes directly from the API:

# Latest-version comments only
curl "https://plsreadme.com/api/comments/abc123def456?view=current"

# Full timeline comments (default API behavior)
curl "https://plsreadme.com/api/comments/abc123def456?view=all"

Viewer links persist the mode in the URL for shareable review context:

• https://plsreadme.com/v/abc123def456?view=current

• https://plsreadme.com/v/abc123def456?view=timeline

To claim a legacy anonymous link into your signed-in account:

curl -X POST https://plsreadme.com/api/auth/claim-link \
  -H "Authorization: Bearer <clerk-session-jwt>" \
  -H "Content-Type: application/json" \
  -d '{"id":"abc123def456","adminToken":"sk_..."}'

MCP (AI Editors)

Current recommendation today:

• use hosted remote MCP with browser login when your client supports it cleanly

• use personal API key fallback when remote auth is unavailable or awkward in that client

• use the local plsreadme-mcp package with PLSREADME_API_KEY for the safest stdio path

Connect your editor to plsreadme and share docs with natural language:

"Share this README as a plsreadme link" "Turn my PRD into a shareable page" "Make these meeting notes into a readable link"

MCP/agent auto-review loop with /versions

For iterative AI writing flows (draft → critique → revise), agents can consume /v/:id/versions as the source of truth:

1. Keep the canonical readable URL (/v/:id) for humans.

2. Poll /v/:id/versions between iterations.

3. Compare current_version to the last reviewed version.

4. If changed, fetch raw_url for the newest version and run review checks.

5. If quality regresses, optionally trigger /v/:id/restore with admin token + owner session.

This gives automation deterministic revision tracking without scraping HTML.

See docs/ai-iteration-versioning.md for a full playbook.

🔌 MCP Setup

Client compatibility matrix

Current as of April 5, 2026:

Hosted Remote Login (supported clients)

Claude Code:

claude mcp add --transport http plsreadme https://plsreadme.com/mcp

Cursor:

{
  "mcpServers": {
    "plsreadme": {
      "url": "https://plsreadme.com/mcp"
    }
  }
}

VS Code:

{
  "servers": {
    "plsreadme": {
      "type": "http",
      "url": "https://plsreadme.com/mcp"
    }
  }
}

Windsurf:

{
  "mcpServers": {
    "plsreadme": {
      "serverUrl": "https://plsreadme.com/mcp"
    }
  }
}

Lifecycle notes:

• access token TTL is about 1 hour

• refresh token TTL is about 30 days

• reconnecting the same client replaces the older grant

• sign out ends the website session but does not automatically revoke an existing editor grant

• use GET /api/auth/mcp-grants and DELETE /api/auth/mcp-grants/:grantId to audit or revoke hosted editor grants

If your client supports browser login, prefer this path. It is the cleanest setup and keeps owned docs tied to your website account automatically.

Hosted Remote API Key fallback

Create a personal API key from https://plsreadme.com/my-links first, then use one of these:

Claude Code:

claude mcp add --transport http \
  --header "Authorization: Bearer $PLSREADME_API_KEY" \
  plsreadme-api https://plsreadme.com/mcp

Cursor:

{
  "mcpServers": {
    "plsreadme-api": {
      "url": "https://plsreadme.com/mcp",
      "headers": {
        "Authorization": "Bearer ${env:PLSREADME_API_KEY}"
      }
    }
  }
}

VS Code:

{
  "inputs": [
    {
      "type": "promptString",
      "id": "plsreadme-api-key",
      "description": "plsreadme personal API key",
      "password": true
    }
  ],
  "servers": {
    "plsreadme-api": {
      "type": "http",
      "url": "https://plsreadme.com/mcp",
      "headers": {
        "Authorization": "Bearer ${input:plsreadme-api-key}"
      }
    }
  }
}

Windsurf:

{
  "mcpServers": {
    "plsreadme-api": {
      "serverUrl": "https://plsreadme.com/mcp",
      "headers": {
        "Authorization": "Bearer ${env:PLSREADME_API_KEY}"
      }
    }
  }
}

Raw remote endpoint users:

curl -i https://plsreadme.com/mcp \
  -H "Authorization: Bearer $PLSREADME_API_KEY"

Local npm fallback

Claude Code:

claude mcp add --transport stdio \
  --env PLSREADME_API_KEY=$PLSREADME_API_KEY \
  plsreadme -- npx -y plsreadme-mcp

Cursor: Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "plsreadme": {
      "command": "npx",
      "args": ["-y", "plsreadme-mcp"],
      "env": {
        "PLSREADME_API_KEY": "${env:PLSREADME_API_KEY}"
      }
    }
  }
}

VS Code: Add to .vscode/mcp.json:

{
  "inputs": [
    {
      "type": "promptString",
      "id": "plsreadme-api-key",
      "description": "plsreadme personal API key",
      "password": true
    }
  ],
  "servers": {
    "plsreadme": {
      "command": "npx",
      "args": ["-y", "plsreadme-mcp"],
      "env": {
        "PLSREADME_API_KEY": "${input:plsreadme-api-key}"
      }
    }
  }
}

Claude Desktop: Add to claude_desktop_config.json:

{
  "mcpServers": {
    "plsreadme": {
      "command": "npx",
      "args": ["-y", "plsreadme-mcp"],
      "env": {
        "PLSREADME_API_KEY": "<paste-your-personal-api-key>"
      }
    }
  }
}

Windsurf: Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "plsreadme": {
      "command": "npx",
      "args": ["-y", "plsreadme-mcp"],
      "env": {
        "PLSREADME_API_KEY": "${env:PLSREADME_API_KEY}"
      }
    }
  }
}

Notes:

• local stdio now expects PLSREADME_API_KEY by default so new docs are owned

• explicit legacy anonymous mode still exists with PLSREADME_ALLOW_ANONYMOUS=1

• create your key from https://plsreadme.com/my-links

Migrating existing anonymous MCP setups

If you already used plsreadme-mcp anonymously:

1. Create a personal API key from /my-links.

2. Add PLSREADME_API_KEY to your MCP client config.

3. Keep PLSREADME_ALLOW_ANONYMOUS=1 only as a temporary compatibility crutch for old workflows.

4. Claim older anonymous links later with /api/auth/claim-link if you still have their admin_token.

The migration rule is simple:

• new automated/editor creates should be owned by default

• anonymous local MCP is now legacy-only and explicit

• the website demo path remains zero-setup even while editor auth gets stricter

add-mcp

npx add-mcp plsreadme-mcp

OpenClaw

clawhub install plsreadme

Docker (for MCP registries / listing checks)

Build and run the stdio MCP server in a clean container:

docker build -t plsreadme-mcp:local .
docker run --rm -i plsreadme-mcp:local

The containerized server uses stdio (no ports, no env vars required).

🛠 MCP Tools

Prompts:

• share-document — Guided flow to share content as a readable link

• refactor-and-share — Uses your AI model to refactor raw text into polished markdown, then shares it

Plain text input? No problem — the MCP auto-structures it into markdown, or you can use the refactor-and-share prompt to leverage your AI's reasoning for a polished result.

.plsreadme Record File

The MCP server tracks your shared documents in a .plsreadme JSON file in your project root. This stores document IDs, URLs, and admin tokens needed for editing and deleting.

⚠️ Add .plsreadme to your .gitignore — it contains admin tokens. The tool will warn you if it's missing.

🏗 Architecture

Built on Cloudflare's edge stack for speed everywhere:

┌─────────────┐     ┌──────────────────┐     ┌─────────┐
│  Web / API  │────▶│  Cloudflare      │────▶│   R2    │
│  MCP Client │     │  Workers (Hono)  │     │ (docs)  │
└─────────────┘     └──────────────────┘     └─────────┘
                           │
                    ┌──────┴──────┐
                    │     D1      │
                    │ (metadata)  │
                    └─────────────┘

• Hono — Lightweight web framework on Workers

• Cloudflare D1 — SQLite at the edge for metadata, comments, analytics

• Cloudflare R2 — Object storage for markdown documents

• Durable Objects — Stateful MCP server endpoint

• Workers AI — Optional fallback for text-to-markdown conversion

📁 Project Structure

plsreadme/
├── worker/
│   ├── index.ts              # Main worker entry
│   ├── auth.ts               # Clerk JWT verification utilities/middleware
│   ├── routes/
│   │   ├── auth.ts           # Auth config/session/protected identity endpoints
│   │   ├── docs.ts           # Document creation & rendering
│   │   ├── comments.ts       # Inline commenting system
│   │   ├── convert.ts        # AI text→markdown conversion
│   │   ├── analytics.ts      # View tracking
│   │   ├── links.ts          # Short link handling
│   │   └── waitlist.ts       # Waitlist & notifications
│   ├── mcp-agent.ts          # Remote MCP server (Durable Object)
│   └── types.ts              # TypeScript types
├── packages/
│   └── mcp/                  # npm package: plsreadme-mcp
│       └── src/index.ts      # MCP server (stdio transport)
├── public/                   # Static assets & landing pages
├── db/
│   └── schema.sql            # D1 database schema
├── docs/
│   ├── ai-iteration-versioning.md # Version timeline/restore patterns for human + agent loops
│   ├── auth-clerk.md              # Auth setup + environment checklist
│   └── runbooks/
│       └── legacy-link-claim-rollout.md
├── skill/
│   └── plsreadme/            # OpenClaw agent skill
└── wrangler.jsonc             # Cloudflare Workers config

🔧 Development

# Install dependencies
npm install

# Run locally
npm run dev

# Deploy
npm run deploy

# Bootstrap schema (fresh local DB)
npm run db:migrate:local

# Audit unapplied migrations (remote + local)
npm run db:migrations:status

# Apply migration files explicitly
npm run db:migrations:apply        # remote
npm run db:migrations:apply:local  # local

Ownership phase migration notes:

• wrangler.jsonc points migrations_dir to db/migrations, so migration status is auditable with explicit list/apply commands.

• Apply db/migrations/004_owner_user_id.sql in existing environments before relying on ownership filters.

• Apply db/migrations/007_doc_attribution_telemetry.sql before relying on doc_create_events or raw_view_count.

• Legacy rows are intentionally backfilled as owner_user_id = NULL (anonymous/public behavior preserved).

• Write routes still run a safe ownership schema ensure step (duplicate-column tolerant) for mixed-env rollout safety.

• See docs/migrations.md for the explicit audit/apply workflow.

MCP package release

plsreadme-mcp is published from packages/mcp by pushing an mcp-v* tag (see .github/workflows/publish-mcp.yml).

cd packages/mcp
npm version patch   # or minor/major
cd ../..
git add packages/mcp/package.json packages/mcp/package-lock.json
VERSION=$(node -p "require('./packages/mcp/package.json').version")
git commit -m "chore(mcp): release v${VERSION}"
git tag "mcp-v${VERSION}"
# push commit + tag from your machine to trigger npm publish workflow

Environment Variables

Start from .env.example and set values in your local/dev/prod environment.

Cloudflare tip: non-sensitive values can live in vars; sensitive values should be set with wrangler secret put.

If OAuth credentials are not configured yet, users can still click Sign in / Use email instead and complete auth through the Clerk-hosted email flow immediately.

Frontend auth shell notes:

• /app.html and /my-links use public/clerk-auth-shell.js (Clerk-native browser SDK wiring).

• Authenticated frontend API calls should read bearer tokens through window.plsreadmeGetAuthToken().

The core sharing functionality still requires zero configuration. Clerk auth, AI conversion, and notifications are opt-in.

For the full auth setup checklist, see docs/auth-clerk.md. For rollout + smoke checks, see docs/runbooks/mcp-auth-rollout-checklist.md.

📊 Limits

🤝 Contributing

Feature ideas? Bug reports? Open an issue.

PRs welcome for bug fixes and improvements.

📄 License

MIT — do whatever you want with it.