Visitors
readwriteWho is on your site, your saved audiences and how retention is going.
- list_website_visitors
- list_audiences
- get_retention
- get_visitor_settings
- create_audience
- sync_audience_to_channel
- add_company_to_crm
- update_visitor_settings
Retention + marketing tools for agents
OpenServe knows who came back to your site, which companies are looking, and who has gone quiet. Connect any MCP agent and it can read all of it, build the audiences, and draft the follow-ups. There is no dashboard to learn. Nothing sends, posts or spends until a person says yes.
$ POST /api/v1/mcp
> tools/list → 53 tools
Retention
One tag on your site. After that the work is your agent's, and the decisions are yours.
Companies visiting, the pages they read, and whether it looked like buying. Your own known people are named when they come back, on any device. Nobody is ever guessed from an IP address.
When a recognised visitor has no email with you yet, your site shows your own opt-in at that moment. Nothing is pre-filled and no address crosses from anyone else's data.
Pricing readers. People who gave you their email and visited this month. Known people who have not been back in 30 days. Your agent saves them with create_audience.
The agent drafts the email or the post in your brand's voice and files it. You approve in Activity. Audiences sync to your own Meta and Google ad accounts, and stay current.
get_retention and get_analytics return list growth, returning-reader rate and what each win-back brought home, as raw rows your agent can reason over.
How you use it
The same session in any MCP client: Claude, Cursor, Codex, n8n, or your own agent.
Eleven companies visited this week. Three read pricing more than once, so I added those to your CRM and drafted a follow-up for each in your brand's voice.
| Company | Visits | Read | Signal | Draft |
|---|---|---|---|---|
| Northpine Labs | 6 | /pricing, /security | buying | waiting on you |
| Driftwave | 4 | /pricing, /demo | buying | waiting on you |
| Copperleaf Data | 3 | /pricing | buying | waiting on you |
| Quillstack | 2 | /blog | reading | none |
The suite
Every family has its own read scope and, where one exists, its own write scope. A tool your grant does not carry is not even listed to the agent.
Who is on your site, your saved audiences and how retention is going.
Companies, deals and follow-ups, in the pipeline you already work.
Campaigns, delivery and the media buyer's proposed moves. Paid changes are proposals.
Lists, sizes and performance. The agent writes the draft; a person sends it.
Connected pages, the traffic they send and post drafts filed for approval.
Read and edit your pages. Publishing is its own owner-granted scope with a daily limit.
Realtime, retention, paths, monetization and interface analytics for your workspace.
Voice, colours, fonts and logo, so everything the agent writes sounds like you.
OpenServe's own specialists as tools: ask one, brief a team, run a flow.
The state of anything the agent filed. Poll it instead of assuming it went out.
Find partner businesses in your collectives and request introductions.
Published articles with your own subscriptions. Reads pay the publisher.
The guarantees
These are enforced by the server on every call. They are not prompt instructions an agent could talk its way around.
There is no send_email and no post_social. An agent drafts, a person approves in Activity, and the approval runs the send.
A budget change or an ad move is a proposal. Only the workspace owner can grant the scope that lets an agent propose one.
Drafts per day, publishes per day and a daily paid budget. A spent cap is a clear tool error the agent can read.
Every row an agent writes shows the grant's name. Activity, change history and Settings all say who did what.
A grant runs as the person who made it, with what their role allows today. Leave the workspace and the grant stops working.
No tool returns another tenant's data, a pooled email or the network's numbers. Consent, Global Privacy Control and opt-outs are honoured for you.
Connect
Make a key in OpenServe under Settings › Organization › Connectors & keys › API & agents, tick the families the agent may use, then paste one of these.
claude mcp add --transport http openserve \
https://www.openserve.com/api/v1/mcp \
--header "Authorization: Bearer os_sk_YOUR_KEY"
{
"mcpServers": {
"openserve": {
"url": "https://www.openserve.com/api/v1/mcp",
"headers": { "Authorization": "Bearer os_sk_YOUR_KEY" }
}
}
}
curl https://www.openserve.com/api/v1/mcp \ -H "Authorization: Bearer os_sk_YOUR_KEY" \ -H "Content-Type: application/json" \ -H "MCP-Protocol-Version: 2025-06-18" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
# discovery, no key needed curl https://www.openserve.com/.well-known/openserve.json # search published content, then read with the person's own access curl https://www.openserve.com/api/v1/open/search \ -H "Authorization: Bearer os_sk_YOUR_KEY" \ -H "Content-Type: application/json" \ -d '{"query":"election"}'
Building an agent product for other people? Skip pasted keys. OpenServe speaks OAuth 2.1 with dynamic client registration, so your users land on a consent screen, pick a workspace and the families you may use, and can revoke you any time. A 401 from the endpoint names the metadata your MCP client needs to find the flow on its own.
For publishers
Agents are becoming readers. OpenServe makes them paying ones, without a second paywall to maintain.
Your journal on OpenServe is already readable by agents through the open gateway. Import an archive, syndicate a feed, or write here.
An agent reads with the access of the person behind it: your member, or a paid OpenServe subscriber if you joined the pool. Everyone else gets your free opening and your offer.
Opt your journal into the pool and you share subscription revenue in proportion to paid readers' reads of your members-only posts. One reader counts once per post, however many agents fetched it.
Every served read carries a signed source receipt and your canonical link. get_content_usage is your statement of agent reads by post.
Plans
There is no separate API product and no per-call credit to count. An agent's work draws on the same plan the person's own work does.
Connect an agent and read your own workspace.
For an owner whose agent does the weekly marketing.
For teams running agents all day.
Prices in USD, billed monthly; annual billing is cheaper. Ad spend goes to your own connected ad accounts and is never charged by an agent. Current limits for each plan are on openserve.com.
Help
No. It can draft an email and file it. A person approves it in OpenServe › Activity, and that approval is what sends. The same holds for social posts and paid changes.
It names companies from business networks, and it names people only when they gave your business their email. It does not buy a person's identity from an IP address, and a network-recognised visitor is shown as recognised, never named.
One script tag. Sites, journals and funnels built on OpenServe carry it already. Consent, Global Privacy Control and opt-outs are handled by the tag.
Anything that speaks the Model Context Protocol over HTTP: Claude, Cursor, Codex, n8n and custom agents. Protocol versions 2025-06-18, 2025-03-26 and 2024-11-05 are supported. Responses are JSON; there is no streaming transport.
Revoke its key or consent in Settings. It stops on the next call. Keys can also carry an expiry date when you make them.
No. When you lack access to an article the agent gets the free opening and the offer links, and shows them to you. Buying is always yours.
Read /llms.txt on this site for the short version and /llms-full.txt for every tool with its scope. Then call tools/list with the key your person gave you; the server lists only what your grant carries.
Make a key, tick the families, paste one line. Your first answer about who visited takes about two minutes.