2dai-mcp-server

2dai-mcp-server

MCP server for 2DAI that enables AI assistants to generate images and videos, upload references, browse the public feed, and manage cloud drive on a 2DAI account.

Category
Visit Server

README

2dai-mcp-server

npm

MCP server for 2DAI — lets Claude, Cursor, Cline, Windsurf and any other MCP host generate images and video, upload references, browse the public feed, and organise your cloud drive on your own 2DAI account.

  • Runs locally over stdio via npx — nothing to install permanently.
  • Authenticates with a 2DAI API key (create one at 2dai.io → Dashboard → Integrations → API keys).
  • Finished images come back as inline previews so the model can see what it made and iterate.

Quick start

Claude Code

claude mcp add 2dai --env TWODAI_API_KEY=2dai_sk_... -- npx -y 2dai-mcp-server

Claude Desktop / Cursor / Cline / Windsurf (JSON config)

{
  "mcpServers": {
    "2dai": {
      "command": "npx",
      "args": ["-y", "2dai-mcp-server"],
      "env": { "TWODAI_API_KEY": "2dai_sk_..." }
    }
  }
}

The key is read from the environment only — never pass it as a tool argument.

Hosted alternative — mcp.2dai.io:800

Every host that speaks remote MCP (Streamable HTTP) can point at our hosted server instead of running npx locally. No install, no Node runtime, no process lifecycle — just a URL and a Bearer key.

Endpoint — https://mcp.2dai.io:800/mcp

Auth — Authorization: Bearer 2dai_sk_... on every request (same key you'd put in TWODAI_API_KEY for the stdio flow).

Claude Desktop / Cursor / Cline / Windsurf (remote-MCP JSON shape — check your host's docs, some hosts still spell the type "remote"):

{
  "mcpServers": {
    "2dai": {
      "type": "streamable-http",
      "url": "https://mcp.2dai.io:800/mcp",
      "headers": { "Authorization": "Bearer 2dai_sk_..." }
    }
  }
}

Claude Code

claude mcp add --transport http 2dai https://mcp.2dai.io:800/mcp \
  --header "Authorization: Bearer 2dai_sk_..."

Both transports (stdio via npx and hosted via HTTP) expose the exact same tools and the exact same behavior. Pick whichever fits: npx when you want zero third-party dependency and full control of the process, hosted when you want zero install and a single URL.

Tools

Tool What it does Scope Spends credit
get_account Account status: credit, tier, key label/scopes/spend cap read no
generate_image Text-to-image (style/quality default to auto) generate yes
generate_with_refs Image from references: face-ref, character-ref, style-transfer, smart-edit (edit refs[0] per the prompt) generate yes
generate_video Animate a still creation into a short clip generate yes
generate_similar Re-run an existing creation ("more like this one") generate yes
generate_wallpaper Expand a creation into a wallpaper dimension (standard, photo, widescreen, ultrawide); quality fixed at Ultra, price follows the dimension generate yes
check_generation Poll a queued generation by queueId read no
cancel_generation Cancel a still-waiting generation (charge refunded); explains itself when it is too late generate no
upload_image Upload a local image / base64 as a reference generate no
download_creation Save the full-resolution asset to disk, or return an inline preview read no
list_creations Page, search, sort and filter the library (folders, trash, activity lenses, smart collections, shared folders, random pick). Rows include nsfwFlagged/nsfwRate so agents can apply their own safeguards read no
browse_feed Page through the public feed (other creators' published work) read no
list_folders Page through the account's folders read no
manage_folder Folder CRUD + favorites, poster, and sidebar groups (create/rename/delete/move-to-group/list-groups) manage no
organise_creation Move / trash / restore / like / unlike one creation; batch-trash / batch-restore up to 100; permanent delete stays one-at-a-time manage no
publish_creation Publish / unpublish a creation on the public feed publish no
get_stats One consolidated stats read: counters, streak, storage, 30/90-day generation volume + spend, top refs/styles/keywords read no
get_wallet_status $2DAI balance, USD credit, effective tier + the four signals it derives from finance no
get_lock Staking-lock status (none / active / expiring-soon) finance no
get_wallet_history Transactions, balance chart, credit chart, or credit sources finance no
get_token_price Cached $2DAI/USD quote with its staleness read no

All wallet tools are read-only — no tool on this server can move money. The finance scope is opt-in and never part of a key's defaults; enable it in the dashboard's key settings when you want an agent to see wallet data.

Least privilege

Scopes are set per key in the dashboard. A coding agent that only generates needs read + generate; add manage for drive organisation, publish only if you want it posting to the public feed, and finance only for wallet reads. You can also set a per-key spend cap — the server reports actionable errors when a cap or scope blocks a call.

Configuration (env)

Variable Default Meaning
TWODAI_API_KEY — (required) Your 2DAI API key
TWODAI_PREVIEWS 1 Attach downscaled inline previews to finished generations
TWODAI_PREVIEW_MAX_SIDE 512 Longest edge of those previews, in px
TWODAI_WAIT_BUDGET_MS 45000 How long a generation call blocks before degrading to a queueId
TWODAI_IDEMPOTENCY_WINDOW_MS 30000 Window in which an identical re-submit is treated as a retry, not a new charge
TWODAI_ALLOW_ANY_PATH 0 Allow file reads/writes outside the working directory
TWODAI_API_BASE production API API origin override (for self-hosted testing)

Notes for agents

  • Generation spends real credit. Check get_account first when unsure, and never generate speculatively.
  • Long generations return a queueId — collect with check_generation; re-submitting would be charged again.
  • Rapid identical submits inside the idempotency window are deduplicated, never double-charged.
  • upload_image / download_creation paths are confined to the working directory unless TWODAI_ALLOW_ANY_PATH=1.

The 2DAI stack

We currently use our Gen 7.1 model, and unlike others, we don't resell model access. All our models originate from the open-source community or our own R&D — we then disassemble, modify, fine-tune and optimize them to align with our legacy and 2DAI❤️ART lines. They also run on our own private cloud network.

Every tool response carries the model version and stack summary (via get_account.platform), so an agent can hand these facts back to a user who asks "what powers you?" without extra plumbing.

License

MIT

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured