1Scope
NimbleAI provides an MCP (Model Context Protocol) server that lets an AI assistant — such as Claude, through Claude Desktop, Claude Code, or Claude.ai Custom Connectors — call a defined set of tools against your NimbleAI account on your behalf.
These terms cover only that integration: what the connected assistant can see, what it can do, and how we handle the data that passes through it. Everything else about using NimbleAI — billing, uploads, the web dashboard — remains governed by our general Terms of Service and Privacy Policy. If the two ever conflict on an MCP-specific point, this page controls.
By enabling or using NimbleAI MCP with any client, you agree to the terms below.
2What each tool can do
MCP tools are scoped narrowly on purpose. Each one does exactly one thing, and none of them can act outside your own account.
| Tool | Access | What it touches |
|---|---|---|
| list_videos() | Read-only | Metadata for your own videos: title, format, viral score, clip count, timestamps. Never the video files themselves. |
| get_job_status(jobId) | Read-only | Processing status of a render job tied to your account. Rejects any job ID that doesn't belong to you. |
| upload_video(url) — planned | Write | Starts a render job using your remaining video credits. Will require explicit confirmation in the client before running. |
| apply_subtitle_style(name) — planned | Write | Changes the subtitle style on a future export. Does not modify clips already delivered. |
Tools marked "planned" are not yet active; this document will be updated before they ship, and the changelog on the MCP page reflects current availability at all times.
3Authentication & tokens
Every MCP tool call is authenticated with a bearer token tied to your NimbleAI account. That token is the equivalent of a password for this integration.
- The token is issued when you connect a client and is never visible to the assistant itself — it's attached to requests by the MCP server, not typed into the conversation.
- Tokens expire automatically and must be refreshed periodically. An expired token simply fails authentication; it does not grant partial or stale access.
- If a client or device storing your token is compromised, revoke it immediately from your account settings (see Section 9) and contact us if you suspect misuse.
- We never ask for your NimbleAI password inside an MCP conversation, and no legitimate tool response will ever request it either.
4Data we share via MCP
MCP tool responses are limited to metadata: titles, statuses, scores, counts, and timestamps. Raw video and audio files never pass through the assistant or through Anthropic's infrastructure as part of an MCP call.
Each tool call is scoped to the authenticated user only — there is no tool that can list, read, or modify another user's data, and row-level access checks are enforced server-side on every request, independent of what the assistant asks for.
Tool calls and their outcomes (not their full payloads) may be logged for reliability, abuse prevention, and debugging, consistent with our general Privacy Policy.
5Third-party AI clients
NimbleAI MCP is built to work with Claude, but MCP is an open protocol — any compliant client can, in principle, connect. We don't control how a given assistant chooses to call our tools, what it says about the results, or how long it retains conversation history.
When you use NimbleAI MCP through Claude specifically, your conversation is also subject to Anthropic's own terms and privacy practices, which are separate from ours. We recommend reviewing them directly at anthropic.com.
We are not responsible for how a third-party AI client interprets, summarizes, or acts on the data our tools return, beyond ensuring that the data itself is accurate at the time it's sent.
6Your responsibilities
- Keep your MCP token and the device or client running it reasonably secure, the same way you would a password.
- Don't share your token with anyone else or embed it in a public repository, config file, or shared environment.
- Use MCP tools only for your own account — attempting to probe, enumerate, or access another user's data through the protocol is a violation of these terms and our general Terms of Service.
- Comply with the usage policies of whichever AI client you connect (for example, Anthropic's usage policies when using Claude).
- Review write-capable tools (once available) before confirming an action — the assistant will ask for confirmation, but you're responsible for what you approve.
7Availability & beta tools
NimbleAI MCP is offered on an as-available basis. Live tools are functional but may occasionally return errors, timeouts, or incomplete data during upstream incidents. Tools marked "Soon" are not guaranteed to ship on any particular date and their eventual behavior may differ from what's currently described.
We may add, modify, deprecate, or remove individual tools at any time, particularly during the current beta phase, with reasonable efforts to reflect changes on the MCP page before they take effect.
8Rate limits & fair use
Tool calls are subject to reasonable rate limits to protect service stability for all users. Automated or scripted use that goes materially beyond typical conversational usage — for example, polling a status endpoint in a tight loop — may be throttled or blocked without prior notice.
Free-tier accounts remain subject to the same credit and clip limits via MCP as they are via the dashboard; MCP is an additional interface to your account, not an additional allowance.
9Revocation & termination
You can disconnect NimbleAI MCP or revoke its token at any time from your account settings. Revocation takes effect immediately for new tool calls; a client may need to be restarted to clear any cached session state.
We may suspend MCP access for an account that violates these terms, our general Terms of Service, or the usage policies of the connected AI client, independently of the rest of the account's standing.
10Liability
NimbleAI MCP is provided without warranty of any kind, express or implied, including as to accuracy, availability, or fitness for a particular purpose, to the maximum extent permitted by law. NimbleAI's aggregate liability arising from your use of the MCP server is limited in the same manner and to the same extent as set out in our general Terms of Service.
We are not liable for actions taken by a third-party AI assistant based on its own interpretation of tool output, or for consequences of a compromised token that resulted from your failure to keep it secure.
11Changes to this policy
We may update this page as MCP tools evolve. Material changes — new write-capable tools, changes to what data is shared, or changes to authentication — will be reflected here with an updated effective date. Continued use of NimbleAI MCP after a change constitutes acceptance of the revised terms.
12Contact
Questions about this policy, a security concern, or a suspected token compromise can be sent to hello@nimbleai-edit.com.