From audience insight to a measured change.
Use an agent to move between the work your team already does: understand an audience, prepare a relevant experience, and assess the result. Each MCP tool has a defined input schema. The tools available to a connection depend on its permissions.
Give an audience a more relevant hero.
Use this documented workflow with the audience and zone IDs returned by your own site.
1. Inspect the audience and page.
Call
get_segmentsandget_zones, thenget_contentfor the selected zone. Check the audience's evidence and the zone's accepted content fields before proposing a change. Segment conversion rates describe observed behavior; they do not prove that a new hero will improve conversion.2. Generate copy for review.
Call
draft_content_variantwith the selectedsegment_id, the zone'szone_key, andvariant_type: "hero_swap". It returns suggested headline, body, and CTA text. This call does not create or publish a content variant. Review its claims and adapt the copy to the zone's field schema before saving.3. Save and publish deliberately.
Check the site's publishing mode first. On a site configured for manual publishing,
create_content_variantorupdate_content_variantsaves the working copy;publish_content_variantsmakes the selected variants' working copies available to visitors. On the default auto mode, visitors can receive saved working-copy changes without a separate publish call. For review before release, use manual publishing and give the drafting agent a site-scoped write token with propose-only enabled, without admin access.4. Read the result after eligible traffic arrives.
Call
get_lift_reportfor the per-variant conversion comparison, andget_scorecardfor the whole site's primary-conversion comparison with its configured holdout. Report the measured window, arm counts, and returned evidence status. A learning or inconclusive result is a reason to keep observing, not a demonstrated win. The whole-site scorecard does not isolate the effect of one hero change.
Example inputs
Replace the uppercase placeholders with values returned by the read tools.
| Tool | Input |
|---|---|
get_segments | {} |
get_zones | {} |
get_content | {"zone_id":"ZONE_ID"} |
draft_content_variant | {"segment_id":"SEGMENT_ID","zone_key":"ZONE_KEY","variant_type":"hero_swap"} |
get_lift_report | {"days":30} |
get_scorecard | {"days":30} |
Read the tool reference or inspect the machine-readable tool manifest for the complete inputs and descriptions.
Start with a useful question.
The agent uses the site's actual tools and records. Permissions determine which actions it can take; publishing mode determines when saved content can reach visitors.
Connect first. Start with reads.
https://axoapp.ai/api/mcp using the setup provided with your token. AXO hosts the server; you do not need to run a local MCP server. Follow the connection guide for authentication and client setup.tools/list. Then read the site's segments and zones. A successful connection confirms tool access; personalization still requires the site tag, suitable zones and content, and a configured measurement goal.Four seller workflows, installed as a plugin.
The axo plugin for Claude packages this connector with skills that join a Salesforce account, contact, lead or opportunity to what that company and its people did on your site. It installs in Claude Code, claude.ai and Cowork, and runs beside Salesforce in Claude. Once connected, the server tells Claude which of the skills your token can run.
The three reporting skills run on a read-scope, read-only token; the drafting skill runs on a write-scope, propose-only token and never publishes. Install the plugin or download axo-plugin-0.3.0.zip (its SHA-256 is on the install page).
Can an AI agent operate website personalization through AXO?
Yes. AXO at axoapp.ai provides a hosted MCP server for reading audience and performance data, generating copy suggestions, managing content, and inspecting holdout results. The agent works through tools with defined inputs and the permissions of its connection.
Does drafting copy change what visitors see?
The draft_content_variant tool returns suggested copy without creating or publishing a content variant. Saving a variant is a separate action. In auto publishing mode, saved working-copy changes can reach visitors immediately. In manual mode, visitors receive published revisions, so your team can review edits before publishing them.
How do permissions and human review work?
Site tokens are scoped to one site. Read access supports inspection; write access adds drafting and editing, including content publishing unless further restricted. Admin access is required to change the site's publishing mode. Read-only and propose-only restrictions can narrow a token further. For review before content goes live, configure manual publishing and give the drafting agent a site-scoped write token with propose-only enabled, without admin access. A write token by itself is not a human approval requirement.
Does connecting an agent prove personalization lift?
No. Connection, content publication, and measurement are separate steps. AXO's holdout reports compare eligible traffic within the returned allocation period. Use the report's evidence status and sample counts; a positive point estimate alone does not establish a win, and too little data does not establish equal performance.
Is there a REST API too?
Yes. AXO documents REST access for profiles and segments through the Pulse API. MCP provides the broader agent tool surface. Choose the documented interface for the operation your integration needs.
Are there ready-made skills for sales teams?
Yes. The axo plugin for Claude bundles the connector reference with four seller skills: pre-call-brief, hot-accounts, explain-exposure and draft-segment-page. They join a Salesforce account, contact, lead or opportunity to AXO website behavior and run beside Salesforce in Claude. The three reporting skills need a read-scope, read-only token; the drafting skill needs a write-scope, propose-only token and never publishes. The plugin installs in Claude Code, claude.ai and Cowork; install steps and the plugin archive with its SHA-256 are at axoapp.ai/docs/reference/mcp#plugin.
