How to Connect a Terminal Coding Agent to Your Outreach Stack
You connect a terminal coding agent to WarmySender by pointing it at one address — `https://warmysender.com/mcp` — and authorising it either by signing in with…
You connect a terminal coding agent to WarmySender by pointing it at one address — https://warmysender.com/mcp — and authorising it either by signing in with your WarmySender account or by handing it an API key from Settings → API Keys. That is the whole pattern; every client below is a variation on those two steps. Once it is connected, your agent can build, launch and manage outreach in plain language, and it still cannot send a single message itself — WarmySender’s scheduler does all the sending, at its own pace, inside your safe per-account limits.
Key takeaways
- One address, two ways in. Point any MCP-capable agent at
https://warmysender.com/mcp, then either sign in with your WarmySender account when the approval screen opens, or export an API key from Settings → API Keys. - The terminal is the least gated route. No marketplace review, no plan tier that hides custom connectors — if the agent runs in your shell and speaks MCP, you can wire it up in a minute.
- Six clients, one pattern. Claude Code, Codex, Cursor, OpenClaw and Hermes Agent are all one command or a few config lines; Meta’s Muse Code reads its servers from a settings file.
- The agent is a planner, not a sender. It can create, start, pause and resume campaigns across cold email, LinkedIn, Instagram and WhatsApp, enrol prospects, verify addresses and tune warmup — but the sending is never its call.
- Access is revocable at any time. Keys can be read-only, and both keys and sign-in sessions are cancelled from Settings → API Keys.
Why a terminal agent is the easiest route
If you already live in a terminal, a terminal agent is the shortest path from “I want an AI to run my outreach” to it actually happening. Chat-app connectors depend on directories, plan tiers and review queues. A terminal agent depends on a config file you own and a command you type. There is nothing between you and the connection.
The second reason is context. Your agent is already sitting in your project, with your notes, your ICP definition, your CSVs and your naming conventions in reach. Adding outreach tools to that session means the agent can go from “here is the segment we discussed” to a drafted, enrolled campaign without you re-explaining anything or exporting a file to a browser tab.
And the third is that the terminal is where iteration happens. Asking “which of my mailboxes look unhealthy?” and then “pause anything sending from those” is two sentences, not two dashboards.
The one thing every agent needs
Everything on this page reduces to a single line of configuration: the server address https://warmysender.com/mcp. What differs between clients is only how you tell them about it, and how you prove who you are.
There are exactly two ways to authorise, and both are equally valid:
| Sign in with your account | API key | |
|---|---|---|
| How it works | The agent opens an approval screen in your browser; you sign in to WarmySender and approve | You create a key in Settings → API Keys (it starts with ws_) and give it to the agent |
| Where the credential lives | Held by the client’s own sign-in flow — nothing is written into your config | In an environment variable or your client’s config, so keep it out of version control |
| Best for | Day-to-day interactive use on your own machine | Scripted or headless setups, and anywhere you want a narrower scope |
| Scoping | Full access as you | Keys can carry read-only permissions if you only want the agent to look, not change |
| Revoking | Settings → API Keys | Settings → API Keys |
A read-only key is an underrated first move. Connect the agent with one, spend a day asking it questions about your campaigns and mailbox health, and only widen the permissions once you like the answers. The full reference for both routes lives in connect your AI agent.
Setting up each client
WarmySender’s positioning list of agents is Claude, ChatGPT, Cursor, Codex, OpenClaw, and Hermes Agent — plus any agent that speaks MCP. Meta’s Muse Code is an additional client that fits the same pattern.
Claude Code
One terminal command registers the server. Then type /mcp inside Claude Code — or run claude mcp login warmysender — and sign in in the browser when it opens. From that point the outreach tools appear alongside everything else in the session, and you can start asking for campaigns in plain English.
Codex
Codex is included with paid ChatGPT plans from Plus upward, which makes it the practical way to drive WarmySender from OpenAI’s stack. Export your key, then:
codex mcp add warmysender \
--url https://warmysender.com/mcp \
--bearer-token-env-var WARMYSENDER_API_KEY
Prefer not to store a key? Add the server without the token flag and run codex mcp login warmysender to sign in instead. Either way the server ends up in ~/.codex/config.toml, so you can check or edit it by hand. The step-by-step version, including the ChatGPT-side context, is in connect WarmySender to ChatGPT and Codex.
Cursor
Cursor gives you the friendliest path: use the Add to Cursor button and it registers the server for you. If you would rather keep the configuration in the repo — which is the right call for a shared project — paste the server into .cursor/mcp.json in your project root. Cursor will ask you to sign in the first time the tools are used.
OpenClaw
Two lines in the terminal, and the second one opens a WarmySender sign-in in your browser. Worth knowing: OpenClaw ignores static headers while sign-in is switched on, so no key is stored anywhere in your setup. If you are security-conscious about credentials on disk, this is the cleanest of the six.
Hermes Agent
Add the server to ~/.hermes/config.yaml, then run /reload-mcp in the chat so Hermes picks up the change without a restart. Sign in when prompted and the tools are live.
Muse Code
Muse Code is Meta’s terminal coding agent, released on 2026-08-05, and it runs on macOS or Linux. It reads its MCP servers from a settings file at ~/.config/muse/settings.json. Two things to know before you edit it: the file needs a schema_version line or Muse Code will reject it outright, and the exact contents matter enough that we keep them in a dedicated guide rather than paraphrasing them here. Add the server as described in connect WarmySender to Muse Code, save the settings file and start Muse Code again. Note that Muse Code is the one client on this list with no browser sign-in option for MCP servers — a key in the settings file is its only route, which is why its guide is written around a key rather than an approval screen.
If you are weighing up whether to run your outreach from Muse Code at all, can Meta Muse run my outreach? answers the scope question, and run cold email outreach with Meta Muse Code walks the whole workflow end to end.
What to ask it once it is connected
The tool catalog is broad, so the fastest way to learn it is to ask for things in ordinary language and watch what the agent reaches for. A starting set, by channel:
Cold email — “Create a three-step campaign for SaaS ops leads in Germany, then show me the sequence before you start it.” Prospects and lists — “Bulk-import this CSV, unenrol anyone already in the Q3 campaign, and add the churned accounts to the suppression list.” Diagnostics — “Give me the report of actions that did not run in the last week and tell me why each one was skipped.” Verification — “How much verification allowance is left? Verify this batch, cancel the run if it looks wrong, and build a list from the deliverable results.” Leads — “Search the B2B leads database for heads of growth at 50-200 person fintechs, save the segment, and export it.” Warmup and health — “Which mailboxes have warmup problems? Show me the health of each one and turn warmup on for the two new domains.” LinkedIn — “Set up a connection campaign for the saved segment, and tell me how the account’s health and throttle look before we start.” WhatsApp and Instagram — “Pause the WhatsApp campaign, check number health, and show me how the Instagram DM account is doing.” Housekeeping — “Check my tracking domains, list saved email templates, and show me the status of anything still running.”
Notice the shape of every one of these: the agent decides, drafts, organises and reports. When something needs to actually leave your account, it stops and hands over.
What it will never do
This is the part worth reading twice, because it is what makes handing an agent the keys reasonable in the first place.
The agent never sends a message, DM or invite itself, and it can never raise a limit. Those two are absolute. Creating or launching a campaign only writes the campaign and hands it to WarmySender’s scheduler. The scheduler paces every email, LinkedIn, Instagram and WhatsApp action inside your safe per-account caps and the gradual warmup ramp — exactly the same way, at exactly the same rate, whether a human clicked the button or an agent asked for it. There is no fast lane for automation, and no instruction you or your agent can give that widens a cap.
Email verification sits under the same roof: it shares the same allowance and the same pacing, so an agent working through a large list cannot burn through your budget faster than you would yourself.
And one thing is deliberately not an agent tool at all: connecting and disconnecting accounts always stays in the app. Adding a mailbox, linking a LinkedIn or Instagram account, removing one — those are manual actions you take yourself, every time. An agent can tell you a mailbox looks unhealthy. It cannot unplug it, and it cannot plug in a new one.
Put together, the worst case for a misunderstood instruction is a campaign you did not want, sitting there for you to pause — not a burst of mail from an account that should have been resting. You can see the full picture of what agents can and cannot touch on the AI agents page.
Frequently asked questions
Do I need an API key, or can I just sign in?
Either works. If you sign in, the agent opens a WarmySender approval screen in your browser and you authorise it there, with nothing written into your config files. If you prefer a key, create one in Settings → API Keys — it starts with ws_ — and give it to the agent, usually through an environment variable. Keys are the better choice for scripted or headless setups, and they can be scoped read-only so the agent can look without changing anything. Both routes are revoked from the same place: Settings → API Keys.
Which terminal agents can connect to WarmySender?
Claude Code, Codex, Cursor, OpenClaw and Hermes Agent are all supported directly, and so is any other agent that speaks MCP. Meta’s Muse Code, a terminal coding agent released on 2026-08-05 for macOS and Linux, also connects by adding the server to its settings file at ~/.config/muse/settings.json. In every case the server address is the same — https://warmysender.com/mcp — and only the registration syntax differs.
Can an AI agent actually send emails or LinkedIn messages for me?
No. An agent can create, update, start, pause and resume campaigns, enrol and unenrol prospects, and configure warmup, but it never sends a message, DM or invite itself and it can never raise a limit. Launching a campaign only writes it and hands it to WarmySender’s scheduler, which paces every email, LinkedIn, Instagram and WhatsApp action inside your safe per-account caps and the gradual warmup ramp, regardless of whether a person or an agent triggered it. Email verification shares the same allowance and pacing.
Can an agent connect a new mailbox or LinkedIn account for me?
No — connecting and disconnecting accounts is the one thing that is deliberately not an agent tool. Adding a mailbox, linking a LinkedIn, Instagram or WhatsApp account, or removing any of them stays a manual action in the WarmySender app. An agent can report on account health, warmup status and throttle for accounts you have already connected, but it cannot add or remove one.
What can an agent actually do once it is connected?
Quite a lot. Cold email campaigns and prospects (create, update, start, pause, resume, enrol, unenrol, bulk-import, manage the suppression list and saved templates, and pull a report of actions that did not run and why), tracking domains, LinkedIn connection and message campaigns plus posts, recruiter jobs and applicants, and account health and throttle, WhatsApp campaigns and number health, Instagram DM campaigns and account health, warmup settings and mailbox health, email verification (single address, batch, remaining allowance, cancelling a run to get unused credits back, and building a list from the results), the B2B leads database (search, save, export), and workspace settings and job status.
Where to go next
Pick your client and follow its guide: connect your AI agent covers the shared setup and both authorisation routes, connect WarmySender to ChatGPT and Codex handles the OpenAI side, and connect WarmySender to Muse Code has the exact settings-file contents for Meta’s agent. When you are ready to actually run something, run cold email outreach with Meta Muse Code is a full worked example, and can Meta Muse run my outreach? sets expectations first. The short version stays the same whichever you choose: one address at warmysender.com/mcp, one sign-in or one key, and a scheduler that keeps every account safe no matter who is driving.