You can run an entire outreach campaign without opening Twain once: create the agent, set up the campaign, add your contacts, and review the messages, all from a Claude conversation. This walkthrough follows a real campaign we ran exactly this way: a single LinkedIn follow-up to around 150 existing users, generated on Ultra.
Before you start: connect the Twain MCP to Claude. Setup takes two minutes, see What is MCP, and how does it work with Twain?. You'll also need a list of contacts as a CSV. Everything else, the agent and the campaign, gets created along the way.
For every supported client (Claude, Claude Code, ChatGPT, Codex, Cursor) and the full list of tools, see the MCP server documentation.
Step 1: Upload your list and set the goal
Attach your CSV to the Claude conversation and give Claude the context it needs in one prompt:
Who is on the list, and how qualified it already is. In our example, the list was fully qualified upfront, which shapes everything that follows: the agent doesn't need to re-qualify anyone, so writing style becomes the most important part of the setup.
What you want to happen: create a new agent, create a campaign, and prepare messages for these contacts.
If your Twain account has several workspaces, Claude will list them and ask which one to use.
Step 2: Create the agent
Ask Claude to create a new agent in Twain. For a pre-qualified list, put your effort into two things:
Writing style rules: how the messages should sound, written as plain instructions.
Blocked words: the words and phrases Twain should never use. If you already keep a blocked-word list in another agent, paste it into the prompt.
Claude creates the agent and returns a link, so you can open it in Twain at any time to check the setup. See How Agent Writing Style Works for what belongs in Rules and Domain Knowledge.
Step 3: Brief the campaign
Next, tell Claude what the campaign should produce. A good brief covers:
Channel and steps. For example: a single LinkedIn follow-up message.
Length. We used 100 to 150 words. For people who already know your product, a slightly longer message that lets each sentence breathe often works better than something ultra-short.
Relationship and tone. Spell out how well you know these people. In our case: friendly and familiar, not too close but not a stranger, written peer-to-peer.
The opener. Open with the hook: why you're reaching out today.
What to say about the product. One or two key points, in your own words.
The CTA. Ours was informational: we added free credits to their account, no question at the end, no meeting ask.
Personalization. One sentence in the body about the recipient: something recent, ideally about the person specifically rather than their employer.
The brief can be written in plain language, exactly as you'd explain it to a colleague. Twain turns it into the campaign setup. For more on how briefs are interpreted, see How Campaign Brief Works.
Step 4: Choose how to generate
Once the list is uploaded and the instructions are complete, Claude doesn't just start spending credits. It asks how you want to handle generation and puts the credit cost of each option right next to it. For our 135-contact Ultra campaign, the choice looked like this:
A few samples now, then import the rest. Claude generates two messages at Ultra (about 20 credits) so you can check the voice and blocked-word compliance before anything else runs. This is the recommended path.
Just bulk-import (0 credits). The contacts land on the campaign without any generation. You batch-generate and review everything yourself in the Twain UI later.
Import and generate the whole list right away. The biggest spend (about 1,350 credits for 135 contacts on Ultra), ready to review in bulk.
Always start with samples. Two or three are enough to judge whether the style rules and the brief land the way you intended, and on Ultra the first message step costs 10 credits per contact, so sampling before a full run protects your budget. See How Modes Work: Base, High, and Ultra for what each mode costs and adds.
Step 5: Review and refine in the Twain UI
Everything Claude created is a regular Twain object. Open the agent and the campaign in the Twain app to see the researched contacts and the generated messages.
Two ways to refine:
Small message edits are easiest directly in the Twain UI: open the message, adjust the wording, done.
Instruction changes (tone, structure, what to mention) can go through Claude: describe what should change, and regenerate the samples.
Step 6: Run the full list and export
Once the samples look right, tell Claude to run the remaining contacts. When generation finishes, export the campaign to your sequencer or CRM the way you always would, either from the Twain UI or by asking Claude to push the finished messages onward.