FlowClick connects both ways through the open MCP standard. An agent step can call tools on any MCP server. And FlowClick itself is an MCP server — roughly 140 tools that let Claude, or any assistant your team already uses, read your data, run your flows, score your models, and operate your processes, with permission.
The mcp step calls a tool on any MCP server — a vendor's, a partner's, your own — and the agent step lets a model use those tools mid-flow. Credentials stay in a Connection; egress is pinned and SSRF-safe like every other outbound call.
Every workspace is an MCP server. Point Claude Desktop, Claude Code, or any MCP client at it with a scoped token and the assistant can operate your business through the same permissioned surface your team and your automations use.
The MCP tools are the platform's real management surface — the same one FlowClick's own agent and its build pipeline use. Grouped roughly:
Records, bulk insert/update/delete, NoSQL documents and collections, key/value, deduplicated uploads, chunk manifests and integrity checks.
Run a flow, run it for each record, cancel it; list, signal, and replay durable processes; complete human tasks; inspect runs and traces.
Create and update any node, import blueprints, audit and fix an app, test it, diff and publish versions, propose / validate / approve / promote changes.
Event-source status, replay, and dead-letter queues; process DLQs; retry jobs; metrics and usage queries; lineage graphs; app health.
Browse, install, upgrade, and publish listings; grant and query entitlements; create sub-accounts and mint their tokens; legal hold and shred.
Assign flows to hybrid runtimes behind your firewall, register and verify custom domains, subscribe outbound webhooks, and manage connections and tokens.
Two more tools matter most to an assistant: get_capabilities returns the live registries of step types, entities, and semantic types, and get_best_practices returns the platform's own rules — so an AI building on FlowClick builds it the way a reviewer would want.
An application's delivery descriptor can declare an agent tools surface. On publish, FlowClick compiles the app's flows into a tool manifest for assistants — alongside its web, PWA, API, push, and offline surfaces — so "give the sales team an app" and "give the sales team's Claude a tool" are the same publish.
An external assistant over MCP is subject to the same controls as FlowClick's own agent: token scopes, role-based permissions, effect budgets, rate limits, and the audit and trace streams. Revoke the token and it stops. AI governance on the Security page →
We'll set up a demo workspace, hand you a scoped token, and let your Claude run the process itself.