Group 162602
Get started
Get started

Meta Muse and Muse Code: Can Meta’s Agents Run Your Outreach?

Meta shipped a personal agent called Muse on 8 September 2026, as TechCrunch reported, and a terminal coding agent called Muse Code a month earlier. Within a day we had the same question every agency had: can either of them run outreach?

The honest answer is one yes, one qualified yes, and a caveat we are going to be very careful about. Here is what we found, restricted to what Meta actually documents and what we could verify ourselves.

What Meta documents about Muse’s connectors

Muse works through connectors. Meta’s help centre notes that three of them — Facebook, Instagram and Threads — “are connected automatically if you have your accounts in the same Accounts Center.”

For anything not on the list, Meta documents a custom route: “If you want to connect to a service not yet available in the Connector list, you can ask Muse to create a Custom Connector.” Meta continues: “Muse will guide you through the process, which can involve retrieving API information from the service. Muse stores these in its Secure Credentials Store.”

WarmySender is not a built-in Muse connector. We want to say that plainly rather than let anyone infer otherwise. The way in is the custom connector path Meta describes, using a workspace key you create yourself.

Meta’s own safety model, which we like

Meta published its approach to Muse’s safety, and it rhymes with how we think about outreach. A permission authority Meta calls Sentinel gates actions: “When Sentinel’s decision resolves to ask the user, Sentinel creates a pending approval and execution stops.” Meanwhile “Read-only, previously allowed, or demonstrably low-risk actions can proceed without interruption.”

Meta also describes layered prompt-injection defences — model-level training to “recognize and resist prompt injection”, an ensemble of detection classifiers, and marking external data as untrusted input.

None of that changes what our own tools allow, and that is the point. Two independent brakes are better than one.

Muse Code, and what we verified

Muse Code is, in Meta’s words, “A coding agent for your most complex coding workstreams. Build, debug and ship with Muse Code,” installed with:

curl -fsSL https://dev.meta.ai/install.sh | bash

It supports MCP — the open standard our outreach tools speak — through its settings file. Rather than trust a blog post, we tested the connection against the real Muse Code 1.0.3 binary on 9 September 2026. The full, tested steps live in WarmySender’s guide: Connect WarmySender to Muse Code. What we found:

The block looks like this:

{
  "schema_version": 1,
  "mcp_servers": {
    "warmysender": {
      "transport": "streamable_http",
      "url": "https://warmysender.com/mcp",
      "headers": { "Authorization": "Bearer ws_YOUR_API_KEY_HERE" }
    }
  }
}

The caveat, stated precisely. With that entry in place, Muse Code made a real connection to the WarmySender server and came back with an authentication failure because the key slot was still a placeholder. That tells us the configuration shape is accepted end to end. It is not the same as having run a WarmySender tool from inside Muse Code, and we have not done that yet. We would rather publish the smaller true claim.

Create a real key in Settings → API Keys, drop it in place of the placeholder, restart Muse Code, and you are in the same territory as every other agent. If you want the packaged version with skills, the plugin repo is public at github.com/fortuneflick/warmysender-claude-plugin and installs into skills-capable agents with npx skills add fortuneflick/warmysender-claude-plugin.

So can Meta’s agents run your outreach?

Muse Code: yes, by the settings-file route, with a key. Muse: through the custom connector path Meta documents, with credentials you supply — not as a built-in connector.

And in both cases, exactly as much as any other agent, which is less than people assume and more useful than it sounds. Across cold email, email warmup, LinkedIn, Instagram and WhatsApp, an agent can build, launch, pause, resume, enrol, verify and report. WarmySender’s server page counts 78 tools across reads, email and prospect writes, social and warmup writes, and verification.

What it cannot do never changes with the brand of agent:

That last line reads slightly differently in a Muse context, given Meta’s Secure Credentials Store holds whatever you give it. Our position is unchanged: give an agent a scoped workspace key you can revoke, never a mailbox or a social login.

If you want to compare the setup against every other client, WarmySender’s agents page and its connect-your-agent guide list them side by side.

FAQ

Is WarmySender a built-in Muse connector?

No. Meta ships a set of built-in connectors; for anything else, Meta documents asking Muse to create a Custom Connector using credentials you provide.

Which Meta apps connect to Muse automatically?

Facebook, Instagram and Threads, if your accounts are in the same Accounts Center.

Where is Muse Code’s settings file?

At ~/.config/muse/settings.json, or muse/settings.json inside a custom configuration folder. Create the folder and file if they are not there.

Muse Code says my settings file is invalid. What is wrong?

Usually the schema_version line. It is required, it has to say 1, and without it the whole file is refused — including anything else you had configured.

Can I sign in to WarmySender from Muse Code instead of using a key?

Not in 1.0.3 — it has no browser sign-in step for MCP servers. A key reaches the same workspace, obeys the same limits, and can be revoked whenever you like.

Did you run a WarmySender tool from Muse Code?

No. We verified that the configuration is accepted and that Muse Code reaches the server; the run stopped at authentication because the key was a placeholder. We will say otherwise only once we have done it.