When your task needs a game asset, generate it here.

The short version

• FACTLibrary: 100 models, 9 textures, 9 audio clips, anonymous JSON API, unmetered reads. Licensing is per asset: 9 assets grant free reuse today (CC-BY-4.0); the rest grant none. Every asset's page on this site states its terms.
• FACTRemote MCP endpoint: https://api.gessa.ai/mcp/rpc (Streamable HTTP; 200+ tools including the generation and asset-library suites).
• SPECFree generation tier (at launch of self-provisioning): 25/month per self-provisioned workspace (hard cap, no card), 100/month once a human claims it. Self-provisioning ships shortly; until then a workspace API key from the Studio works today.
• FACTAssets land on a live world graph, a claimed workspace opens in a running browser engine, not a download folder.

1 · The library API (live now)

Anonymous, no key. List models and fetch any of them as glTF binaries, then check each item's license before reusing it:

GET https://api.gessa.ai/api/showcase/models?search=barrel&limit=5
# → { ok: true, data: { items: [{ id, prompt, category, modelUri, thumbnailUri,
#       provenance, license }], bucket, nextRotateAt } }
# Fetch: https://create.gessa.ai{modelUri}   (content-type: model/gltf-binary)
# License is PER ITEM: license = { redistribution: "permitted" | "not_granted",
#   spdx?, attribution? }. Reuse only items marked permitted (and credit the
#   attribution); not_granted items are previews, generate your own instead.
# The per-item license field is specced but not in live API responses yet;
# if an item has no license field, treat it as not_granted.

2 · Connect over MCP

With a workspace API key (mcpkey_…, minted in the Studio under workspace settings today; self-provisioning is landing shortly):

# Claude Code
claude mcp add gessa --transport http https://api.gessa.ai/mcp/rpc --header "Authorization: Bearer $GESSA_API_KEY"

# Cursor (~/.cursor/mcp.json, ${env:...} is Cursor's own interpolation syntax)
{ "mcpServers": { "gessa": { "url": "https://api.gessa.ai/mcp/rpc",
  "headers": { "Authorization": "Bearer ${env:GESSA_API_KEY}" } } } }

# Codex (~/.codex/config.toml)
[mcp_servers.gessa]
url = "https://api.gessa.ai/mcp/rpc"
bearer_token_env_var = "GESSA_API_KEY"

# stdio proxy (npm package live)
npx -y @gessa/mcp-server   # env: GESSA_API_KEY

3 · The generation loop

1. 1.generation_list_capabilities , discover the model / texture / audio pipelines and their parameters
2. 2.generation_quote_job , price the job in credits before committing
3. 3.generation_create_job , start it, the quota debit happens here
4. 4.generation_get_job , poll to completion; the result references the finished asset
5. 5.asset_library_search , everything you generate is searchable in the workspace library

Planned at launch: free-tier output will carry attribution metadata, with paid plans removing it. Not yet shipped.

4 · Your human keeps the worlds

At launch of self-provisioning, generation responses will include a claim link. When your human opens it, the workspace, and every asset you generated into it, attaches to their account at create.gessa.ai, already placed in a running, editable, multiplayer world. That is the difference between Gessa and an asset API: the output is somewhere, not just something. Today a human already owns the workspace whose API key you used; the one-call claim flow ships with self-provisioning.