Let your coding agent do the growth work.
Claude Code, Cursor, or any MCP client can list your sealoutreach agents, create one, read leads, and draft replies in your voice — over a real MCP server. It never posts anything on the spot. Sending stays off unless you give a token that one permission.
What an agent knows
Five plain-language fields, whether you write them in the dashboard or your coding agent sets them via create_agent: who counts as a fit, what a post worth replying to looks like, what to never chase, how to write the reply, and the voice to write it in.
You stay in control of sending
A connected coding agent can read leads, create agents, qualify posts, and draft replies. Sending is a separate permission, off by default. With it, the agent can approve a draft, which queues it to post a few minutes later, never on the spot, so you can see it waiting in your dashboard first.
What MCP can do
List and create agents, read leads, qualify posts, draft replies. With the send permission: queue an approved draft to post after a few minutes.
What only you can do
Post a reply right away, or switch an agent to fully automatic. Both run on your account and only from the dashboard.
Connect in minutes
A personal access token, not a browser OAuth flow — generate one in your dashboard and paste it into your client's config.
Generate a token
In your dashboard sidebar, click Connect an agent, name the token, and pick its scopes.
Add the server
Point your MCP client at the hosted endpoint below with the token as a bearer credential. For Claude Code:
For a client that reads a JSON config instead:
Call a tool
Ask your client to list your agents, create one, or draft a reply for a lead.
| Protocol | Streamable HTTP MCP |
|---|---|
| Hosted endpoint | https://sealoutreach.com/api/v1/mcp |
| Authentication | Bearer personal access token, generated in your dashboard, revocable anytime |
| Permissions | agents:read, agents:write, leads:read, runs:qualify, runs:draft, and runs:send (off by default; queues with a delay, never posts immediately) |
Agent-readable facts
No browser automation and no instant posting. A token only reaches one workspace, and you can revoke it the moment you no longer want a client connected.
What your agent can do
Six tools, each behind its own permission. Grant only what a given client needs.
list_agentsList the connected workspace's agents.
create_agentCreate a new lead-finding agent — name, targeting criteria, and voice.
list_leadsRead the leads collected for one of the workspace's agents.
qualify_leadScore a lead against its agent's targeting criteria.
draft_replyDraft a reply in the agent's voice for a qualified lead. Does not send it.
queue_sendApprove a drafted reply. It's queued to post after a few minutes, never immediately. Needs the runs:send permission, which is off by default, and a plan that can post.