Meta Muse and Muse Code: What They Mean for Outbound Sales Teams

Meta shipped two very different agents weeks apart, and only one of them is a realistic way to drive an outreach stack from a prompt today: Muse Code, the terminal coding agent, which reads MCP servers from a config file and can therefore talk to WarmySender the same way Claude, ChatGPT, Cursor, Codex, OpenClaw and Hermes Agent already do. Muse, the consumer personal agent announced on 8 September 2026, is a different product with a different job — it browses, books and handles errands on your behalf, and WarmySender is not one of its built-in connectors. Whichever agent you point at your workspace, the guardrail is the same and it is not optional: the agent never sends a message, DM or invite itself, and it can never raise a limit.

Key takeaways

• Muse Code is the one that can touch your outreach stack today. It runs in your terminal, it reads MCP servers from ~/.config/muse/settings.json, and WarmySender publishes a hosted MCP server, so the connection is a config entry plus an API key.
• Muse is a consumer agent, not an outbound tool. It connects to services through built-in connectors, user-supplied API credentials, or a browser — and WarmySender is not a built-in Muse connector today.
• The agent never sends anything. Creating or launching a campaign only writes it and hands it to WarmySender’s scheduler, which paces every email, LinkedIn, Instagram and WhatsApp action inside safe per-account caps and the gradual warmup ramp — the same pacing whether a human or an agent triggered it.
• An agent can never raise a limit. There is no tool for it. Email verification runs on the same allowance and the same pacing as everything else.
• Connecting and disconnecting accounts stays in the app. That is the one thing that is deliberately not an agent tool, no matter which agent you use.

What is Meta Muse?

Muse is Meta’s personal AI agent, announced on 8 September 2026 as “a secure, private personal AI agent that proactively helps with people’s goals and suggests ideas.” Meta’s framing is that it “doesn’t just answer questions, it actually does the work” — it “can handle tasks, like sending an email or booking travel,” it “can open a browser, fill out forms, and negotiate on their behalf,” and it “keeps working after people close the app.”

The architecture is the interesting part for anyone thinking about handing an agent real credentials. Muse runs on what Meta calls the Muse Secure VM — a dedicated secure computer with its own browser, described as where “the data and credentials for any service a person connects are securely stored.” A separate Sentinel agent runs on that same machine, and per Meta, “Nothing Muse does reaches the internet unless the Sentinel approves it.” Meta also states that “Muse has no visibility into people’s passwords or payment methods,” that users “choose which apps Muse connects to and exactly how much access it gets,” and that they “can change access or disconnect a service whenever they want.”

It is rolling out in the US on iOS, Android and muse.ai, with AI glasses noted as coming soon. There is a free tier for most use, plus paid subscription plans above that.

On how it reaches services it wasn’t shipped with, TechCrunch’s coverage of the launch is specific: Muse ships with built-in connectors for several services, and “If a service the user wants isn’t available but offers a public API, Muse can set up a connection using credentials the user provides. When no API is available, Muse can access the service via the browser instead.” The same coverage adds that Muse is reachable through WhatsApp chats as well as the app and the web — a surface Meta’s own announcement does not list, so treat it as reported rather than documented.

What is Muse Code, and how is it different?

Muse Code is a different animal entirely. Released on 5 August 2026 and described by Meta AI Research as “a terminal coding agent powered by Muse Spark 1.2,” it “takes on complex software engineering tasks across large repositories: planning changes, writing code, and validating the results.” It runs on macOS or Linux and is labelled beta.

Two design details matter beyond coding. It uses “async background agents” that “remain active throughout each session, rather than being spawned for individual tasks,” and TechCrunch’s 5 August 2026 report adds that it “fans out to separate sub-agents working in parallel in isolated worktrees.” It also keeps “a local event log in which every model call, tool run, approval, and edit is appended,” which Meta says makes the runtime “replay-exact and restart-safe.”

For an outbound team, the practical point is narrower and more useful than any of that: Muse Code is a terminal agent that reads MCP servers from a config file. Anything that speaks MCP is a tool it can call — including a WarmySender workspace.

Which one can run my outreach today?

Muse Code, via MCP. WarmySender runs a hosted MCP server at https://warmysender.com/mcp, and Muse Code reads its MCP servers from ~/.config/muse/settings.json. You add the server there and connect with a WarmySender API key created in Settings → API Keys (it starts with ws_). There is no sign-in screen to click through: Muse Code has no browser sign-in flow for MCP servers, so the key in the settings file is the whole authorisation step. The step-by-step is in Connect WarmySender to Muse Code, with the outreach-specific walkthrough in Run cold email outreach with Meta Muse Code.

Muse is a maybe, and an honest one. WarmySender is not a built-in Muse connector today. What the public reporting says is that when a service isn’t built in but offers a public API, Muse can set up a connection using credentials the user provides — and WarmySender does have a public API with keys you create yourself. We have not tested a Muse connection to WarmySender and are not going to claim otherwise. If you want to try it, start from Can Meta Muse run my outreach?, and treat it as an experiment rather than a supported path.

Neither agent replaces the six we position around — Claude, ChatGPT, Cursor, Codex, OpenClaw and Hermes Agent, plus any agent that speaks MCP. Muse Code is simply one more agent that can drive the same tools. The full picture is on the AI agents page and the MCP page.

Muse vs Muse Code at a glance

What Muse can and cannot do with a WarmySender workspace

Whatever agent is holding the key, the tool surface is the same. An agent connected to a WarmySender workspace can:

• Cold email — create, update, start, pause and resume campaigns; enroll prospects; bulk-import prospects; manage the suppression list and saved email templates; pull the skipped-actions report.
• LinkedIn — build connection and message campaigns, handle posts, work with recruiter jobs and applicants, and read account health.
• WhatsApp and Instagram — WhatsApp campaigns and number health; Instagram DM campaigns and account health.
• Deliverability plumbing — tracking domains, warmup settings and mailbox health.
• Email verification — single and batch checks, allowance, cancel-and-refund, and building a list from the results.
• Data and admin — searching, saving and exporting from the B2B leads database, plus workspace settings and job status.

And here is the list of things no agent can do, on any platform, by design:

• It cannot send. Not an email, not a LinkedIn message or invite, not an Instagram DM, not a WhatsApp message. Creating or launching a campaign writes the campaign and hands it to WarmySender’s scheduler. The scheduler does the sending.
• It cannot raise a limit. There is no tool for it, so no prompt — clever, urgent, or accidental — can produce a bigger daily cap. Email verification draws on the same allowance and the same pacing as everything else.
• It cannot connect or disconnect an account. Attaching a mailbox, a LinkedIn account, a WhatsApp number or an Instagram profile happens in the app, with you in the browser. That boundary is deliberate and it is the one thing that is not an agent tool.

The guardrail that matters most

Outbound punishes speed. A mailbox that suddenly triples its volume, a LinkedIn account that fires off a burst of invites, a brand-new domain that sends like an established one — those are how accounts get throttled and burned, and no amount of agent intelligence makes them safe.

So WarmySender puts the pacing where it cannot be argued with. The scheduler paces every email, LinkedIn, Instagram and WhatsApp action inside safe per-account caps and the gradual warmup ramp, regardless of whether a human or an agent triggered it. An agent asked to “launch this to everyone today” produces exactly the same send pattern as a human clicking Launch: the campaign is written, handed over, and then metered out at a rate the accounts can carry. Email verification shares that same allowance and pacing.

That is why “the agent never sends” is not a limitation we apologise for. It is the reason it is reasonable to let an agent build campaigns at all.

How to get started

1. Create an API key. In WarmySender, go to Settings → API Keys and create one. It starts with ws_.
2. Add the MCP server. For Muse Code, add https://warmysender.com/mcp to ~/.config/muse/settings.json and connect with your API key — Muse Code has no browser sign-in step for MCP servers, so the key is the whole authorisation. Full instructions: Connect WarmySender to Muse Code.
3. Start with reads. Ask for mailbox health, warmup status or a campaign summary before you ask for anything that writes. It is the fastest way to confirm the connection and to see how the agent describes your workspace back to you.
4. Then let it build. Draft a campaign, enroll a segment, verify a list, adjust warmup — all in plain language. Then open the app and look at what it wrote before you launch.
5. Using a different agent? The generic path is Connect your AI agent, which covers Claude, ChatGPT, Cursor, Codex, OpenClaw, Hermes Agent and anything else that speaks MCP.
6. Want the map rather than the manual? Connect an AI agent has a one-paste setup block for every supported client and is where you create your key, and the WarmySender MCP server lists every tool an agent can reach.

Frequently asked questions

Can Meta Muse run my WarmySender campaigns?

Not through a built-in connector — WarmySender is not one of Muse’s built-in integrations today. Public reporting on Muse says that when a service isn’t built in but offers a public API, Muse can set up a connection using credentials the user provides, and WarmySender does offer a public API with keys you create in Settings → API Keys. WarmySender has not tested a Muse connection and makes no claim that it works. The terminal agent Muse Code, which reads MCP servers from a config file, is the supported path today.

How do I connect Muse Code to WarmySender?

Muse Code reads MCP servers from ~/.config/muse/settings.json. Add WarmySender’s hosted MCP server at https://warmysender.com/mcp and connect with a WarmySender API key — created in Settings → API Keys, beginning with ws_. Muse Code offers no browser sign-in for MCP servers, so the key is the only route, and the settings file needs a schema_version line saying 1 or Muse Code refuses the whole file. The full walkthrough is at warmysender.com/documentation/connect-warmysender-to-muse-code.

Can an AI agent send emails or LinkedIn messages on my behalf?

No. No agent connected to WarmySender ever sends a message, DM or invite itself, and none can raise a limit. Creating or launching a campaign only writes it and hands it to WarmySender’s scheduler, which paces every email, LinkedIn, Instagram and WhatsApp action inside safe per-account caps and the gradual warmup ramp — identical pacing whether a human or an agent triggered the work. Email verification shares the same allowance and pacing.

What is the difference between Meta Muse and Muse Code?

Muse is a consumer personal AI agent announced on 8 September 2026 that handles tasks like sending an email or booking travel, runs on a dedicated secure machine with its own browser, and is rolling out in the US on iOS, Android and muse.ai, with AI glasses noted as coming soon. Muse Code is a terminal coding agent released on 5 August 2026 that plans changes, writes code and validates results across large repositories on macOS or Linux, and is currently in beta. They are separate products with separate purposes.

Do I have to let an agent connect my mailbox or LinkedIn account?

No, and you cannot — connecting and disconnecting accounts is deliberately not an agent tool. Attaching a mailbox, LinkedIn account, WhatsApp number or Instagram profile always happens in the WarmySender app with you in the browser. An agent can read account health and configure warmup, but the connection itself stays under your direct control.

If you want the practical setup rather than the landscape view, start with Connect WarmySender to Muse Code and Run cold email outreach with Meta Muse Code. If you are curious about the consumer agent instead, read Can Meta Muse run my outreach? before you experiment. And if your agent of choice is something else entirely, Connect your AI agent covers the rest.