AI assistants (MCP)
Connect ChatGPT, Claude, Gemini CLI, Grok, Claude Code, Codex, Cursor, Windsurf or VS Code to Zyhap Marketing. The assistant plans, writes and schedules posts, reads analytics, and runs campaigns and ads, always as the person who connected it and never beyond their permissions.
The server
One Streamable HTTP endpoint. Most people connect from the AI assistants page in Marketing, which shows the steps for each assistant.
https://mcp.zyhap.com/mcp
Name the server after the module (for example marketing) in your assistant's config.
Signing in (OAuth 2.1)
ChatGPT, Claude, Claude Code and Codex connect with just the URL. The server supports Dynamic Client Registration (RFC 7591), so the assistant registers itself and sends the person to sign in. There they choose the workspaces it may work in and press Allow connection. In ChatGPT, add it from Plugins, Add, Add custom MCP server, with Authentication set to OAuth.
- Metadata:
/.well-known/oauth-authorization-serverand/.well-known/oauth-protected-resource - Registration:
POST /public/v1/oauth/register - Clients without a secret must use PKCE with
S256 - Access tokens last a day and refresh automatically
Connection keys
For clients that take a header, create a connection key on the AI assistants page: read and write or read only, the workspaces it may use, and 30 days, 90 days or 1 year. It is shown once.
{
"mcpServers": {
"marketing": {
"url": "https://mcp.zyhap.com/mcp",
"headers": { "Authorization": "Bearer <your-connection-key>" }
}
}
}Who may connect
A person needs marketing.ai_assistants.connect in their Marketing permission preset, in each workspace they connect. marketing.ai_assistants.manage lets someone see and disconnect every assistant in the branch, with its activity. Company owners can always connect.
Every tool call is checked against the person's own permissions and the workspaces they chose. Launching, pausing, resuming and ending campaigns and ads, and budget changes, need the same permissions as the buttons in the app, and are marked destructive so assistants confirm first. Posts still follow each workspace's approval rules and are labelled with the assistant that made them.
Tools
Workspaces and accounts
get_accessible_social_workspaces, get_connected_social_platforms, get_brand_voices
Posts
get_posts, get_scheduled_posts, schedule_post, update_post, optimize_post_for_platform, generate_hashtags
Approvals
get_pending_approval_posts, approve_post, reject_post
Analytics
get_workspace_analytics, get_connected_platform_analytics, get_post_performance
Campaigns
list_campaigns, get_campaign_performance, get_campaign_launch_checklist, create_campaign_draft, create_campaign_from_template, submit_campaign_for_approval, launch_campaign, pause_campaign, resume_campaign, end_campaign, set_campaign_budget, apply_campaign_budget_suggestion
Ads
get_ads_overview, list_ad_campaigns, get_ad_campaign, list_ad_accounts, preview_ad, create_ad_draft, update_ad_creative, import_media, launch_ad_campaign, pause_ad_campaign, resume_ad_campaign, end_ad_campaign
Money goes in and comes back in whole units. Ask the assistant to list its tools for the full set.
Building your own integration instead? See the REST API.