Your agents, callable from any tool.
The pioneer move: creative work exposed as a service. One MCP endpoint, an API key, and 24 tools that turn your agents into programmable team members in any workflow.
One config block, every MCP client
Create an API key in the app under Settings, then point Claude, Cursor, or your own MCP client at the endpoint. There is no SDK to install: it is a standard streamable HTTP MCP server with bearer auth.
Keys are shown once at creation, then stored only as a hash. Copy it immediately; revoke and rotate anytime. Up to ten active keys per account.
{
"mcpServers": {
"vilva": {
"type": "http",
"url": "https://api.vilva.ai/mcp",
"headers": {
"Authorization": "Bearer vk_live_..."
}
}
}
}24 tools, four groups
Generate directly when you know what you want, or hand the whole brief to an agent. Everything reads and writes the same workspaces you see in the app.
Generation
list_modelsgenerate_imagegenerate_videogenerate_voiceovergenerate_musicprocess_imagecheck_generationget_assetImage waits inline and returns when ready; video and music run async, then you poll. Every gen anchors to a workspace node so it shows up in the app.
Agent
agent_create_runlist_agentsagent_get_runagent_get_planagent_respondagent_send_messageagent_cancel_runagent_list_runsThe full handoff. Discover agents with list_agents, then start a run with a brief, addressed by agentId or agentSlug: platform agents, agent teams, or your own custom agents. Omit both and Verve, the default Vilva agent, takes it. Read the plan and its estimated credits, then approve, amend, or reject before anything generates.
Workspace
list_workspacescreate_workspaceget_workspace_overviewget_nodeupdate_nodelist_workspace_assetsRead and write the canvas the agent works on. update_node deep-merges, exactly like the app, so external tools and the UI stay in sync.
Account
get_credit_balancelist_api_usageCheck the balance before a big run and audit what each call spent.
Brief to plan to estimate to run
The same approval flow as the app, driven over the API. A run never spends without a plan and an estimate first, and budget ceilings with circuit breakers mean a runaway run pauses rather than burning credits.
Budget-safe by design: set a ceiling and the run pauses at the limit. Concurrent API runs are capped per account so a loop can never fan out unbounded.
Send a brief
Call agent_create_run with a brief in plan-first mode. Vilva auto-creates a workspace named for the run if you do not pass one.
Receive a plan and a cost estimate
The run pauses at present_plan. Call agent_get_plan to read the proposed stages and the estimated credits before a single model fires.
Approve, amend, or reject
Reply with agent_respond. Approve to proceed, amend to adjust the plan, or reject to stop. The decision is yours, programmatically.
Vilva executes and you poll
On approval the run executes server-side, no open tab required. Poll agent_get_run for progress and pull finished assets with get_asset.
Same credits as the app. No new billing to wire up, no per-seat tax on automation.
Creative on tap
A campaign tool orders creative the moment a brief lands: agent_create_run with the brief, auto-approve under a set budget, then attach the finished video and voiceover to the campaign. Creative production becomes one step in a flow that used to wait days for a hand-off.
A team bot takes a request in a thread, fires a run, posts the plan and its credit estimate back for a human to approve with a click, then drops the rendered assets into the channel. The whole studio briefs an expert agent from where they already work.
A nightly job refreshes ad variants: it calls generate_image and generate_video against the latest copy, checks the credit balance first, and fails loudly if a budget would be exceeded. Fresh creative ships on a schedule with no one in the loop until review.
Wire creative into your stack.
Create an API key in the app and connect in two minutes. Pay-as-you-go on app credits.