Bring Cordial into any AI assistant.

Your assistant reasons over the live account — 56+ tools across 17 domains — everything from audience inspection to message performance, AI-generated insights, and semantic search across past campaigns. Reads are live today; write actions, with a human approving every one, land soon.

Quickstart → Browse all tools

Claude Desktop config

{
  "mcpServers": {
    "cordial": {
      "type": "http",
      "url": "https://mcp.cordial.io/mcp",
      "auth": "oauth"
    }
  }
}

~/Library/Application Support/Claude/claude_desktop_config.json

Tools

56+

Domains

17

Account data

Live, not exported

Writes

With approval · soon

Why MCP, not just an API key

A wrapped API isn't context.

Point an agent at raw endpoints and it can make calls — it just can't make the right ones. It doesn't know what 'high-intent' means for your brand, which account it's working in, or who the customer on the other end actually is. So it guesses, and it guesses confidently. The job of the MCP server isn't to expose endpoints; it's to give the agent the context to use them correctly — the brand, the account, and the customer.

Brand context

What “VIP,” “win-back,” or “high-intent” actually means here — your channels, your naming, your conventions. Semantic search lets the agent find the audience or flow you mean by describing it, not by guessing an internal ID. And composite tools like get_audience_health return the numbers the way a marketer reasons about them — not three raw rows to stitch together.

Account context

Every tool call is account-aware. The agent always knows which account, which scopes, and which channels are live — it never operates on the wrong tenant, and never makes you repeat which brand you meant. The years of orchestration and audience logic already in that account come along for the ride.

Customer context

Not an anonymous row — a real person, with identity resolved across email, SMS, and push, plus the profile and engagement history behind them. The agent reasons about the customer's actual journey before it proposes the next message.

What it is

A Cordial account, exposed to your AI.

The Cordial MCP server is a hosted Model Context Protocol endpoint. Any MCP-compatible AI client — Claude Desktop, Cursor, Claude Code, your own — connects once and gains a structured, account-aware view of your Cordial data.

The tools mirror the structure of the platform itself: identity, audiences, messages, orchestrations, content, analytics, and so on. Every list endpoint paginates. Composite tools bundle multi-step lookups (like get_audience_health) so agents stay within token budget.

Reads are live today. Write actions — launching a send, editing a contact — are coming to MCP with a human approving each one before it fires: the agent proposes, you commit. (Need writes right now? They're on the REST API.)

Example session

→ tool_call: whoami
← { username: "alex@brand.com", account: "acme-prod", scopes: ["read"] }

→ tool_call: get_account_overview_document
← { channels: ["email","sms","push"], subscribers: 8.4M, ... }

→ tool_call: list_audiences {"query":"high intent"}
← { audiences: [{ id:"aud_8f2a", name:"High-intent · 30d", count:1284902 }, ...] }

→ tool_call: get_audience_health {"id":"aud_8f2a","period":30}
← { detail: {...}, count: 1284902, trend: [...] }

Loyalty tier 3+ has grown 14.2% over the last 30 days, now at 198,455 contacts — see attached trend chart.

One tool for every concept in your account.

Each tool is small, typed, and composable. Composite tools chain common multi-step lookups so agents stay efficient.

Identity

• whoami — Current user identity and account info

Account Context

• get_account_overview_document — Comprehensive account context: channels, attributes, audiences, config
• describe_contact_schema — All contact attribute definitions with types and validation rules
• get_account_supplements — Overview of supplement (data) tables in the account
• get_account_products_config — Product catalog configuration overview
• describe_account_events — Event definitions (system and custom) for behavior tracking
• get_account_audience_samples — Sample audiences with criteria and counts

Audiences

• list_audiences — List audiences with search and pagination
• get_audience — Audience detail: criteria, translation, attributes
• get_audience_count — Current contact count for a saved audience
• estimate_audience — Estimate count for ad-hoc criteria without saving
• describe_audience — Full description: criteria, translation, and count

Messages

• list_messages — Batch sends or automated templates
• list_child_sends — Individual sends of an automated template
• get_message — Message detail with content and performance dashboard
• export_message_stats — Export performance stats as CSV

Analytics

• get_account_overview — Subscriber counts and per-channel stats
• get_message_report — Delivery and engagement rates by message
• get_audience_trends — Audience size trends over time
• list_insight_reports — AI-generated batch message insight reports
• get_insight_report — Executive summary and recommendations

Content

• list_html_includes — Reusable HTML content snippets
• get_html_include — HTML include by ID or key
• search_images — Search uploaded images by query and tags
• get_image — Image details by ID

Orchestrations

• list_orchestrations — Journey/lifecycle automations
• get_orchestration — Detail with DAG and Mermaid flowchart

Sculpt

• list_sculpt_blocks — Reusable email content components
• get_sculpt_block — Block detail with content schema
• list_sculpt_templates — Full email layouts built from blocks
• get_sculpt_template — Template detail with body content

Reports

• list_saved_reports — Saved message reports
• run_saved_report — Execute and return current data
• describe_report_export_columns — Available export columns for a channel
• describe_report_entities — Entity values for a report filter dimension

Channels

• list_channel_types — Configured messaging channels
• list_message_transports — Delivery providers for channels
• get_message_transport — Transport detail by ID

Data Operations

• list_data_automations — Triggered and recurring automations
• get_data_automation — Triggers and schedule
• list_data_batches — Batch data jobs
• get_data_batch — Batch job detail with transform config

Supplements

• list_supplements — Supplement (custom data) tables
• get_supplement — Supplement table detail by key
• list_supplement_records — Records in a supplement table

Products & Tags

• list_products — Products in the catalog
• get_product — Product details by ID
• list_tags — Tags for organizing content assets
• get_tag — Tag details with asset counts

Semantic Search

• search_audience_examples — Semantic search for audiences
• search_orchestration_examples — Semantic search for orchestrations

Composite

• get_audience_health — Detail + count + size trends
• get_message_performance — Dashboard stats + report context
• get_recent_account_activity — Latest jobs and recent batch messages
• compare_insight_reports — Compare multiple insight reports
• explain_message_performance — Full dashboard stats for analysis

You stay in control

The agent proposes. You commit.

Today the agent reads — it can explore every audience, message, and report in your account to reason about what to do next. Write actions are coming soon, and every one will pass through a human approval step before it fires: the agent stages the change, you see exactly what it'll do, and nothing ships until you say go. You get an autonomous assistant with a hand on the controls.

• Reads live now — audiences, messages, orchestrations, analytics, all account-scoped.
• Writes coming soon — every mutation gated behind explicit human approval.
• One browser sign-in, scoped to a single account; the server holds only short-lived tokens, never your password.

Cursor config

// .cursor/mcp.json
{
  "mcpServers": {
    "cordial": {
      "url": "https://mcp.cordial.io/mcp",
      "transport": "http"
    }
  }
}

Quickstart

Three steps to a Cordial-aware agent.

1. Add the MCP server

   Paste the config snippet into Claude Desktop, Cursor, or any MCP-compatible client. Point it at https://mcp.cordial.io/mcp.

2. Sign in once

   The client opens an OAuth flow in your browser. Sign in with your Cordial credentials; the token is stored in your client.

3. Start asking

   Ask your assistant about audiences, message performance, or orchestrations. Tools are called automatically as needed.

Ready to wire your AI into Cordial?

We'll provision an MCP endpoint scoped to your account, plus invite the team members who need access.

Get an MCP endpoint → Need writes today? See the REST API