Your dialer, inside your AI. Built by talking, safe by default.

Tell Claude, ChatGPT or any MCP client what you want. DylrDev builds it in Five9, Genesys Cloud, NiCE CXone, RingCX, Twilio, Telnyx and Amazon Connect. Reads run instantly, writes ask in your AI, and risky actions wait for you.

Create a free account Read the setup docs

One URL per connection: https://dylr.dev/mcp/your-connection

A real Claude conversation, building through DylrDev. Recorded, not mocked up.

As seen on

Editorial feature in IBTimes. Newswire releases on AP News, Yahoo Finance, Business Insider, Benzinga and The Globe and Mail. Articles in Open Source For You and International Business Journal. Interview on the Politech podcast.

Works with the dialer you already run. Seven platforms live today.

DylrDev is independent and not affiliated with or endorsed by these companies. Logos are shown to identify compatibility.

Tool counts and setup for each platform

What it does, exactly. No servers to run, no code to write.

Ask in plain English

Your AI turns the sentence into real tool calls against your platform.

Every tool has a tier

The server enforces it. A clever prompt cannot talk its way past.

It shows you what it built

IVRs and flows come back as a diagram you can check before anything goes live.

Risky actions wait for you

Approve the exact call, or arm the connection for 15, 30 or 60 minutes. Then it locks again.

Everything is logged

Every call, allowed or blocked, with redacted arguments. Export it as CSV.

One URL, any AI

Paste it into the assistant you already use. OAuth or a CLI token, your choice.

How it works, step by step

Try these first. Copy one, paste it into your AI.

Build me an IVR with business hours
Why isn't my campaign dialing?
Draw my main Twilio IVR as a diagram
Show last week's dispositions

Three tiers, enforced on our server. A confirm button in a chat is not a security boundary.

Tier	What happens	Examples
read	Runs right away. Nothing changes in your platform.	`list_campaigns` `get_agent` `run_report`
write	Your AI client asks you to confirm, and the tool shows the exact change first.	`update_skill` `create_disposition`
needs arming	Refused until you approve that exact call or arm the connection in the web app. Approvals expire in 15 minutes.	`stop_campaign` `publish_flow` `upload_call_list`

Arming lasts 15, 30 or 60 minutes, is checked on every call, and ends the moment you disarm or revoke. Deleting anything, placing calls and spending money always need arming.

Tiers, arming, approvals and presets

Proof over promises. Every build shipped as a real clip first.

LinkedIn impressions, last 12 months: 269k
Open-source MCP servers: 6
Tools across live platforms: 250+

Ryan, you've really put CXone through its paces here! The fact it built the IVR, checked its own work and then handed control back to a human before going live is pretty cool to see.

NiCE, official LinkedIn page, in the comments on Ryan's open-source cxone-mcp. A comment on that project, not an endorsement of DylrDev.

RSBuilt by Ryan Shatzkamer on LinkedIn (opens in a new tab). More at ryan-shatzkamer.com (opens in a new tab).

We store as little as possible. And we can show you what we did.

Encrypted at rest: Each connection gets its own key. Credentials are sealed with it, and that key is wrapped with a master key held as a secret. The app only ever shows the last 4 characters.What the app shows:api_secret ••••••••••••a91f
Least privilege first: The setup guides walk you through a dedicated API user with minimal scope. Your AI can only do what that user can do.
Out-of-band arming: The unlock lives in the web app behind your sign-in and a fresh password check. A model cannot click it.
A real audit log: Every call with redacted arguments, the client that made it and the outcome. Kept 90 days, exportable.
Revoke means delete: One click hard-deletes the credentials, tokens, approvals and arming state for that connection.
Pinned to your instance: We remember which host you connected and refuse to talk to any other. The app, sign-in and anything near your credentials run no analytics and no outside scripts, apart from Cloudflare's bot check on the sign-up, password reset and invite forms.

Read the security page

Questions people ask. Short answers, no fine print.

Is it really free?

Yes. DylrDev is a solo project run by Dylr LLC. To keep it fair, each connection has limits: 120 calls per minute, 5,000 per day, 25 items per bulk call, and a 15 second upstream timeout.

Do you see my conversations with the assistant?

No. We host MCP and nothing else. We only see the tool calls your assistant sends us, and we log those with arguments redacted so you can audit them.

What if my assistant gets prompt-injected?

Reads stay reads. Writes make your client ask you. Tools that need arming are refused by our server unless the connection is armed or that exact call was approved in the web app.

What do you store?

Your email and username, a salted hash of your password, connection labels, encrypted credentials, hashed tokens and an audit trail. Audit rows lose their arguments and are purged after 90 days. On the public pages, and only if you allow it in the cookie bar, we also count visits and clicks to see how people find DylrDev. The privacy page has the full list.

Which assistants work?

Any client that speaks MCP over Streamable HTTP. Claude and ChatGPT sign in with OAuth. Claude Code, Cursor, Gemini CLI, Copilot and scripts can use a CLI token for one connection. See the docs.

My dialer is not listed. Can I ask for it?

Yes. Sign in and choose Request a dialer. Other people can add their vote, and the most requested platforms get built first.

Who is behind this?

Ryan Shatzkamer, through Dylr LLC in Florida. Say hello at ryan@dylr.dev.

Connect your dialer. It takes about five minutes.

Create a free account Read the docs