What’s new
The last two weeks were about reach and revenue: taking the MCP servers you build on FlowCP to more places, letting you get paid for them, and giving you more ways to run and observe them. Here’s a tour of what shipped, grouped by what it lets you do.
Ship your server as a chat bot
You can now distribute a published MCP server as a hosted chat bot in Slack, Discord, Telegram, or WhatsApp. People @mentionmention the bot in a channel (or DM it) and it answers in-thread using your server’s tools — the same agentic loop as the Playground and the embeddable chat widget, now living where your team already works.
Crucially, the bot runs in each chat user’s own account, not one shared identity. The first time someone invokes a user_oauth tool, the bot privately DMs them a sign-in link; after they connect, it acts as them. Those sign-ins now refresh automatically, so users aren’t re-prompted every time an access token expires. And when a tool needs confirmation — anything destructive — the bot posts Approve / Reject buttons right in the thread, and only the person who started the turn can act on them.
→ Distribute as a chat bot
Get paid for your MCP server
FlowCP now supports agentic payments end to end. You can charge for tool calls using the x402 standard, settled through a real Coinbase CDP facilitator, with the platform fee billed via Stripe. A new Monetization area gives you earnings analytics and payout UX, and you can test the whole paid flow safely in the Playground with a test wallet before you go live.
→ Agentic payments (x402) · Monetize your MCP server
Run it your way, export and self-host
If you’d rather own the runtime, you now can. Export a server as an installable Claude Code plugin, or run the self-hosted Docker image with plan-based feature gating and signed licenses. Best of all, an exported runtime no longer freezes at export time: with snapshot sync, it polls FlowCP and hot-swaps your latest published config in memory — no restart, no redeploy — while failing closed to the last good config if anything goes wrong. Secrets are never pulled; your upstream credential stays where you put it.
→ Export & self-host your MCP server · Export as a Claude Code plugin · Self-host the platform
Build and test faster
The Playground and embeddable chat were rebuilt on a cleaner component set, and the Playground gained Inspector-style direct testing surfaces plus live streaming of MCP logging and progress notifications — you see what your server is doing as it happens.
You can also now author tools by hand. A new Manual app source lets you define a tool’s method, path, and schema directly — no OpenAPI document required — and test-call it inline. We also added client-side field selection so callers can trim large responses, and fixed a Bubble edge case where list results could be truncated or collapsed to empty objects.
→ Chat-test in the Playground
See more, get help faster
Teams on Enterprise can now stream execution logs to DataDog, Splunk, or a generic HTTP endpoint from a new Log streaming page — the same privacy-safe projection as the Execution logs view (never tokens, headers, or bodies), with per-destination filtering and SSRF-validated targets.
And when you need a hand, there’s now an in-app Support form in the account menu: pick a category, describe the issue, optionally attach an image or video, and it emails a ticket to our team attributed to your workspace.
→ Stream logs to external sinks · Get support
