slideless-mcp

MCP server for Slideless. Wraps the Slideless HTTP API as Model Context Protocol tools so users can list, share, upload, and manage HTML presentations from any MCP host (Claude Desktop, claude.ai, ChatGPT desktop, Cursor, etc.) without installing the slideless CLI.

Hosted on Vercel (Next.js App Router + mcp-handler) at https://mcp.slideless.ai/mcp. Listed on the official MCP registry as ai.slideless/mcp.

Architecture

Stateless Next.js route handler — streamable-HTTP transport, no session store. A per-request McpServer is built on each /mcp call and the user's Authorization header is forwarded verbatim to the Slideless Cloud Functions in europe-west1. No database, no secrets to rotate — the user's token never leaves the request.

Claude / ChatGPT  →  mcp.slideless.ai/mcp  (Vercel / Next.js)
                          │
                          ▼
                 europe-west1-slideless-ai.cloudfunctions.net

Related MCP server: marp-agent-mcp

Authentication

The server is an OAuth 2.1 resource server. It does not host the auth flow itself — it advertises app.slideless.ai as the authorization server and forwards tokens to the Cloud Functions, which validate them. Two header styles are accepted, both forwarded verbatim:

• OAuth (recommended) — hosts that speak OAuth (Claude Desktop, claude.ai) discover the flow automatically: a no-auth request to /mcp returns 401 with a WWW-Authenticate header pointing at /.well-known/oauth-protected-resource (RFC 9728), which points at app.slideless.ai. The user signs in / signs up and consents in-browser; no key is ever pasted. New API keys are minted invisibly by the authorization server.

• Static API key — hosts that support custom headers (Cursor, Claude Code) can send Authorization: Bearer cko_… directly.

Use as a connector

1. In Claude Desktop / claude.ai / ChatGPT: add a custom connector with URL https://mcp.slideless.ai/mcp.

2. The host discovers OAuth and opens the Slideless sign-in. Authorize, and you're connected.

3. (Static-key hosts only) Get a cko_… key at https://app.slideless.ai → Settings → API Keys, and set it as the Authorization header.

Tools

Local development

pnpm install
pnpm dev            # next dev → http://localhost:8787

Test with the MCP Inspector:

npx @modelcontextprotocol/inspector
# Transport: Streamable HTTP
# URL: http://localhost:8787/mcp
# Header: Authorization: Bearer cko_<your-key>

Or with raw curl (initialize handshake):

curl -s http://localhost:8787/mcp -X POST \
  -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" \
  -H "Authorization: Bearer cko_<your-key>" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"probe","version":"0"}}}'

Type checking: pnpm typecheck.

Deploy

Pushed to main → Vercel auto-deploys production (project codika/slideless-mcp). Manual deploy:

vercel deploy --prod --scope codika

Rate limiting is enforced at the edge by Vercel WAF rules (managed via the vercel firewall CLI), matching path starts-with /mcp:

• mcp-per-ip → 60 req / 10s, keyed by IP

• mcp-per-key → 600 req / 60s, keyed by the Authorization header (so each API key / OAuth token gets its own bucket)

Both return 429 when exceeded, before the function runs. Recreate with:

vercel firewall rules add "mcp-per-ip" --condition '{"type":"path","op":"pre","value":"/mcp"}' \
  --action rate_limit --rate-limit-window 10 --rate-limit-requests 60 --rate-limit-keys ip --yes
vercel firewall rules add "mcp-per-key" --condition '{"type":"path","op":"pre","value":"/mcp"}' \
  --action rate_limit --rate-limit-window 60 --rate-limit-requests 600 --rate-limit-keys header:authorization --yes
vercel firewall publish --yes

Source layout

app/
├── layout.tsx                                  # minimal root layout
├── page.tsx                                    # landing page (/)
├── [transport]/route.ts                        # MCP endpoint (/mcp); per-request client + 401 challenge
└── .well-known/oauth-protected-resource/route.ts  # RFC 9728 metadata → app.slideless.ai
src/
├── config.ts             # base URL, server identity (SEP-973 branding), instructions
├── http.ts               # OAuth metadata, CORS, 401 challenge builders
├── server.ts             # registerAllTools entry point
├── slidelessClient.ts    # typed fetch wrapper around the Cloud Functions
├── types.ts              # wire shapes (mirrors slideless-app types/)
├── errors.ts             # SlidelessApiError + wrapToolErrors
└── tools/
    ├── identity.ts       # slideless_whoami
    ├── presentations.ts  # list / get / versions / download / delete
    ├── upload.ts         # upload_html / upload_files (3-step orchestration)
    ├── sharing.ts        # tokens, version mode, unshare, email
    ├── collaborators.ts  # invite / uninvite / list
    └── marketplace.ts    # search / get / remix / publish / star / unstar
server.json               # MCP registry metadata (ai.slideless/mcp)

Related repos

• slideless-app — backend Cloud Functions this server proxies (also hosts the OAuth 2.1 authorization server)

• slideless-cli — npm CLI that uses the same API

• slideless-plugin — Claude Code plugin (companion authoring + upload skills)