Dora is an agentic client portal. Dora, the AI account manager inside it, turns client asks into clear briefs and gets the work signed off. Your team, or your own AI agents, do the work in between.
- 1
A client asks
In their portal or by email, however they like. Dora asks what's missing and writes the brief.
- 2
Your agent picks it up
Run /dora:work_on_request in Claude Code, or hand it to Codex. It reads the brief, the files, and the client's library.
- 3
The client signs off
The work lands for review with a preview link, and the client approves it in one click.
An agent on the client side
Dora answers clients, asks for what's missing, and turns "fix the thing" into a brief before anyone starts. She works from the client's library and rules, with only the access of the person she's helping.
Your agents on the work
Claude, ChatGPT, Claude Code, and Codex connect over MCP and get a request's brief, files, and the client's library in one call. The same call brings the client's rules, brand profile, and similar work already delivered.
Work handed back for sign-off
An agent hands work over with a note and a preview link. The client approves it in one click, or it approves itself after the days you choose. If they ask for changes, the note comes back on the request.
Changes ask first
Reads run right away. Anything that changes a request, sends something to a client, or changes access is marked, so you're asked first. Every tool says what it does, so Claude and ChatGPT know when to check.
You can see who did what
Every action an agent takes is labeled with the tool that did it, like "via Claude Code". Agents act as the person who connected them, and any app can be disconnected in Settings.
Work starts when a client asks
Webhooks fire on new requests, comments, and approvals, so your own setup can start the moment work comes in. Each event is signed, and failed deliveries are retried.
A client portal where AI agents do part of the job, not just a place to keep files and messages. In Dora, an AI account manager handles the client side, turning requests into briefs and getting work signed off, and your own agents can pick up the work over MCP or the API. People and agents work from the same requests, files, and client rules.
No. Most teams start with people doing the work and Dora handling everything around it. Connect an agent when you want one, for one kind of request or one client. Nothing changes for your clients either way: work arrives for review the same way, with a note and a preview link, whoever did it.
Claude, ChatGPT, Claude Code, and Codex, plus anything else that connects over MCP or the API. Claude and ChatGPT connect with a sign-in and Dora's consent screen. In Claude Code, /dora:work_on_request runs a request from brief to handover. Scripts and your own agents can use an API key, which can be read-only or limited to one client.
Anything that sends something to a client or changes a request is marked, so your AI tool asks you first. You decide what it's allowed to do. An agent also acts as the person who connected it, so it can never reach further than they can, and everything it does is labeled with the tool it came from.
Their portal, and Dora. Work your agents hand over arrives for review like any other, with a note, a preview link, and files, and the client approves it or asks for changes in one click. Internal notes an agent leaves stay with your team, the same as anyone's.
From one call. get_request_context returns the request and its thread, its files, what that kind of request needs, the client's rules and decisions, their agreement and brand profile, related library items, and similar requests already delivered with what was handed over. Dora's connection also tells the agent to follow the client's rules and work the way she does.
The client reviews it like any other work. They can ask for changes with a note, and the request goes back to your team, or your agent, with that note on it. Nothing an agent hands over is final until the client approves it or your auto-approve wait runs out, so keep that wait longer while you're trying agents out.
Yes. Add a webhook and Dora POSTs request events to your URL: new requests, status changes, comments, submitted work, approvals, change requests, and estimates. Each one is signed with a secret you see once, and failed deliveries are retried after 1, 10, and 60 minutes. Your setup can start an agent the moment a client asks.
Yes. MCP isn't only for your team. A client can connect Claude or ChatGPT and ask what's waiting on them, open a request, or search their library. Their connection acts as them, so it sees their own workspace and nothing else, and team tools like handing over work aren't open to them.
No. Your first client is free, forever, and each client after that is $10 a month, with MCP and the API on every plan. Your own agents working through Dora's MCP have no usage limit. Only asking Dora herself for her take counts against your company's monthly Dora allowance.
Add your website and your portal is set up in about a minute.
- Free for your first client
- No card needed
- Set up in a minute
- Works with Claude and ChatGPT