Quick start
Connect your agent with OAuth and find your first product-matched references.
Connect Nabit to an agent that supports remote MCP servers and browser-based OAuth. Your agent can find products, request matching video references and work with saved references inside your authorized workspace.
Included with Basic
Basic includes MCP access and image/video generation on the website. MCP does not expose image or video generation. Website and MCP actions share the workspace's Credits and spending controls.
1. Prepare your workspace
Sign in to Nabit. Choose the workspace you want to connect and make sure it has an active subscription, trial or eligible invitation membership. A Credit pack alone does not unlock recommendations or MCP. Ask an owner or admin for access if you cannot spend Credits.
2. Connect with OAuth
In your agent's MCP settings, add a remote server named Nabit:
https://nabit.ai/mcpOn the beta environment, use https://beta.nabit.ai/mcp and a beta account.
Use Streamable HTTP, then sign in through the browser and review the workspace consent screen. No API key is required. Client menus differ; use a client that supports this transport and OAuth. For the existing Codex integration, follow Use in Codex.
Approve the intended workspace, return to your agent, and ask it to call get_connection_status. Check connected, accessAllowed and requiresWorkspaceAuthorization. A successful connection can still have accessAllowed: false when the workspace needs an eligible plan.
3. Find your product
Try this prompt:
Check my Nabit connection, then list my products. Ask me which product to use.
Do not start a paid recommendation request yet.The agent uses list_products and get_product. Use the returned product ID; do not invent one. If needed, preview_product can read a public product page without saving it. Confirm the facts before asking the agent to create_product.
4. Request references
For the product I selected, check my available Nabit Credits.
Explain the cost, then request one batch of matching video references.get_account_usage reports the current balance and recommendation pricing. A new batch reserves up to 8 Credits, then settles at 1 Credit per delivered Hook, up to 8. Read Plans & Credits before enabling automation.
request_recommendations can return a queued or running batch. Keep its snapshotId and poll get_recommendation_batch with that ID. Queued work is not a completed result. Once ready, inspect the references or use render_video_recommendations in a client that supports MCP Apps; viewing links remain available in other clients.
To ask for more, use next_recommendation_batch with the current snapshotId. Retrying that request with the same ID avoids advancing twice. A new batch can spend more Credits; polling an already settled batch does not charge it again.
Direct HTTP access
The implemented /v1 interface supports recommendation batches and existing product trends with the same workspace OAuth authorization. Use the HTTP API reference for routes, parameters, polling and billing. API-key authentication is deferred to milestone 4; MCP continues to use OAuth. These routes require authentication and are not mounted in the documentation-only local preview.