Which AI agents can run your outreach?
Any agent that speaks MCP can connect to WarmySender — Claude, ChatGPT, Cursor, Codex, OpenClaw, and Hermes Agent among them — but what each app is allowed to do differs by app and by plan, and that is where most setups stall. The table below is the honest version, with each vendor's own documentation linked so you can check us. We also publish one-paste setup for Claude Desktop, Claude Code, Grok Bot, Windsurf and Zed.
Whichever you pick, two things never change. Your agent connects to your own workspace at https://warmysender.com/mcp, using either a sign-in or a key that starts with ws_. And your agent never sends a message itself and can never raise a limit — WarmySender's scheduler paces every email, invite and DM inside safe caps and the gradual ramp no matter which app is driving.
What does "speaks MCP" mean?
MCP — the Model Context Protocol — is an open standard for connecting an AI app to outside tools through one common interface. WarmySender publishes an MCP server, so any app that speaks the protocol can drive it without anybody building a special integration first. That is why the list below is open-ended: an agent released next month that speaks MCP will work on day one.
The protocol also builds in the consent step you would want: an app has to ask before it uses a tool on your behalf, and several of the apps below make you confirm each action that changes something.
Which apps can connect today?
| App | How you connect it | Sign-in or key | Plan notes |
|---|---|---|---|
| Claude (web and desktop) | Customize → Connectors → Add custom connector, paste the WarmySender address | Sign in with your WarmySender account — no key needed | Custom connectors are available on Free, Pro, Max, Team and Enterprise; Free is limited to one. On Team and Enterprise an owner adds it for the organisation first, then each member connects |
| Claude Code | One command in your terminal registers the server | Key, or sign-in | Works wherever Claude Code does |
| ChatGPT (web) | Turn on developer mode under Settings → Security and login, then add a custom connector | Sign-in or key, depending on your workspace | The one to read carefully. Full connector support — the write actions campaigns need — is a Business, Enterprise or Edu capability; a personal Pro account is limited to reading; write actions always ask you to confirm. On Plus, use Codex instead |
| ChatGPT (desktop app) | Settings → MCP servers → Add server, choose the streamable HTTP type, paste the address, restart | Sign-in (an Authenticate button) or key | Newer and still rolling out — if the screen is not there, use Codex |
| Codex (CLI and IDE) | codex mcp add warmysender --url https://warmysender.com/mcp, or edit its config file |
Either — codex mcp login warmysender signs in, or pass your key |
Included with every paid ChatGPT plan from Plus up. The reliable ChatGPT-side route |
| Cursor | Add WarmySender to .cursor/mcp.json in your project, or the same file in your home folder for every project |
Key in the headers, or sign-in | No plan restriction we are aware of |
| OpenClaw | openclaw mcp add with the address, or openclaw mcp set with the address and your key |
Either — openclaw mcp login warmysender signs in |
Self-hosted, so it is your own setup that decides |
| Hermes Agent (Nous Research) | hermes mcp add warmysender --url https://warmysender.com/mcp, or add the block to its config file |
Key in the headers, or the sign-in setting | No plan restriction we are aware of |
| Grok Bot | Open the connectors page, create a custom connector, paste the address and your key as the bearer token | Key | Connectors install account-wide, so it is available in any chat |
| Windsurf (Cascade) | The MCP panel in Cascade, or its config file | Key in the headers, or sign-in | Supports remote servers directly |
| Zed | Settings → AI → MCP Servers → Add Remote Server, or its settings file | Key in the headers — leave the header out and Zed offers the sign-in instead | No plan restriction we are aware of |
| Anything else that speaks MCP | Point it at https://warmysender.com/mcp | Either | If it speaks the protocol, it works — no integration needed on our side |
Sources, all checked on 6 September 2026: Claude custom connectors · ChatGPT developer mode · Codex MCP setup · Cursor MCP · OpenClaw MCP · Hermes Agent MCP · Windsurf Cascade MCP · Zed MCP. These are other companies' products and they change quickly — if your screen does not match, trust your screen.
Which plans allow an agent to actually change things?
For every app on the list except ChatGPT on the web, the answer is simply "whichever plan lets you add a custom MCP server at all". ChatGPT is the exception worth planning around: adding your own connector sits behind developer mode, the write actions a campaign needs are reserved for its Business, Enterprise and Edu workspaces, and a personal Pro account can only read. If you are one person on ChatGPT Plus, the answer is Codex — it is included with your plan and it does everything. The full walkthrough is in connect WarmySender to ChatGPT and Codex.
You can also decide this from our side rather than theirs: create a key in Settings → API Keys with only read permissions and any agent becomes read-only, whatever its own plan allows.
Sign-in or key — which should I use?
- Sign-in suits your own laptop and one person: you point the app at the WarmySender address, approve on the screen that opens, and there is no key to look after.
- A key suits terminals, servers and shared setups where nobody is there to click an approval — and it is the only route that lets you choose exactly which permissions the agent gets, including read-only.
Both reach the same workspace and both stay inside the same safe sending limits. Either can be revoked in Settings → API Keys, and access stops immediately.
MCP or the API — when does each make sense?
Use MCP when a person is in the loop: you are talking to an agent, describing what you want, and reading the answers. Use the API when software is in the loop: your own product, a scheduled script, a workflow tool that needs the same call to behave the same way every night. They reach the same capabilities and the same safety limits, so it is genuinely a question of who is asking, not of what is possible. Builders can start at integrations and API.
Frequently asked questions
Which AI agents can run my outreach?
Claude, ChatGPT, Cursor, Codex, OpenClaw, and Hermes Agent — plus any other agent that speaks MCP, including Claude Desktop, Claude Code, Grok Bot, Windsurf and Zed, which we publish one-paste setup for. They all connect to the same WarmySender address, and the same safety rules apply to every one of them: the agent never sends a message itself and can never raise a limit.
What does "speaks MCP" mean?
MCP is an open standard for connecting an AI app to outside tools through one common interface. WarmySender publishes an MCP server, so any app that speaks the protocol can drive it with no special integration built first — which is why an agent released next month will work on day one.
Which plans let an agent actually create and launch campaigns?
For every app except ChatGPT on the web, any plan that lets you add a custom MCP server at all. ChatGPT is the exception: adding a connector sits behind developer mode, the write actions a campaign needs are reserved for Business, Enterprise and Edu workspaces, and a personal Pro account can only read. On ChatGPT Plus, use Codex — it is included with your plan and it does everything.
Can I use a free Claude plan?
Yes, with one caveat: Claude supports custom connectors on Free, Pro, Max, Team and Enterprise, and a Free account is limited to a single custom connector. On Team and Enterprise an owner adds the connector for the organisation first, and then each member connects it individually in their own settings.
Should I connect with a sign-in or an API key?
Sign-in is simplest on your own machine — point the app at the WarmySender address and approve on the screen that opens, with no key to look after. A key is the right choice for terminals, servers and shared setups, and it is the only route that lets you choose exactly which permissions the agent gets, including read-only. Both reach the same workspace and both can be revoked instantly.
Should I use MCP or the API?
MCP when a person is in the loop and talking to an agent; the API when software is in the loop — your own product, a nightly script, a workflow tool. They reach the same capabilities under the same safety limits, so it is a question of who is asking rather than what is possible.
Does my agent get more or less power depending on which app I use?
No. What differs is whether the app lets you connect a custom server at all, and whether it makes you confirm each change. Once connected, every agent reaches the same capabilities in your workspace, under the same limits — and none of them can send a message, raise a limit, or connect an account.
Related reading
Using an agent we have not listed? Email [email protected] and we will help you point it at the right address.