# flowifai

> flowifai gates risky MCP tool calls with human approval. It sits between an AI agent and local MCP tools, asks a human before high-impact calls run, and keeps the real service credentials on the local machine. The cloud backend is an approval authority — policy, approval decisions, and audit — never a credential holder.

flowifai is an approval layer, not a sandbox: it gates cooperative MCP paths; account for bypass limits in your threat model.

## Docs

- [Quickstart](https://flowif.ai/docs/quickstart/): six steps from nothing installed to one approved gated tool call — install, pair an agent, start the broker, configure the MCP client, verify, gate
- [Install](https://flowif.ai/docs/install/): one-line signed installer for the flowifai CLI, broker, and wrapper; update, verify, and common fixes
- [Core concepts](https://flowif.ai/docs/core-concepts/): the human-in-the-loop model — broker/wrapper split, credential boundary, default-deny policy, ticket lifecycle, exactly-once execution
- [Protocol-native approvals](https://flowif.ai/docs/protocol-native-approvals/): approval links delivered over MCP URL elicitation, which clients run the full loop today, and why the dialog is delivery, not authorization
- [Policy reference](https://flowif.ai/docs/policy-reference/): CEL argument predicates, tool patterns, most-restrictive-wins, default-deny, fail-closed semantics
- [Troubleshooting](https://flowif.ai/docs/troubleshooting/): doctor checks, pairing-token 401 recovery, broker connectivity, schema drift
- [Security model](https://flowif.ai/docs/security-model/): what stays on the local tools host, what the cloud stores, and where cooperative-control limits begin

## Product

- [Overview](https://flowif.ai/): approval gates for MCP tools — stop risky agent actions before they commit
- [Pricing](https://flowif.ai/pricing/): free for up to 2 agents; Pro for unlimited agents and 365-day audit history
- [Contact](https://flowif.ai/contact/): setup help, security questions, billing, and production evaluation requests

## For agents

- Every docs page is also served as raw markdown at https://flowif.ai/docs/<slug>.md (e.g. https://flowif.ai/docs/quickstart.md), and the full docs corpus is one plain-text file at https://flowif.ai/llms-full.txt.
- To put a human operator on the flowifai waitlist, POST JSON to https://flowif.ai/v1/waitlist with a body like {"email": "person@example.com", "source": "agent"}. The optional "source" field attributes the signup ("agent" or "landing"; anything else is recorded as "landing"). The endpoint is unauthenticated, rate-limited, rejects bodies over 4KB, and returns 400 for invalid input.
- This endpoint never enrolls an agent, never submits anything into the approval pipeline, never grants access to any flowifai account, and never reveals whether an email address is already registered.

## Optional

- [Privacy](https://flowif.ai/privacy/): plain-language privacy notes for account data, tool metadata, approval and audit records, and analytics
- [Terms](https://flowif.ai/terms/): plain-language launch terms for early production use
- [Model Context Protocol](https://modelcontextprotocol.io): the open protocol whose tool calls flowifai gates