Short answer: Abmatic AI now ships a native MCP server at mcp.abmatic.ai. It gives Claude, Claude Code, and any MCP-compatible agent live, read-only access to your ABM data: accounts, contacts, campaigns, de-anonymized visitors, conversions, opportunities, and analytics. 28 tools, one URL, five-minute setup. Book a demo and we will connect it to a live account during the call.
Quick answer
What is the Abmatic AI MCP server? It is a remote Model Context Protocol server at
https://mcp.abmatic.ai/mcpthat lets AI agents read your Abmatic AI account. MCP is the open standard, introduced by Anthropic, that gives AI assistants a uniform way to use external tools and data (modelcontextprotocol.io). Instead of exporting CSVs or copying dashboard numbers into a chat, you connect once and your agent queries live data on demand. Every tool is a read: agents can look, never touch.
Key takeaways
- 28 read-only tools cover every customer-facing object in Abmatic AI: accounts, contacts, campaigns, revealed companies, visitor sessions, conversions, opportunities, target groups, engagement, reports, dashboards, meetings, and CRM sync status.
- Works with the clients your team already uses: the claude.ai custom connector, Claude Code, and agents built on the Claude Agent SDK or Managed Agents. Any MCP client that supports Streamable HTTP can connect.
- Read-only by design. There are no write or mutation tools. An agent with access can never edit a campaign, alter your CRM sync, or change a setting.
- Account-scoped security. Access uses your Abmatic AI API key as a bearer credential. Keys are revocable from the dashboard with immediate effect, rate limited to 100 requests per minute, and every request is audit logged.
- Built for agents, not just chat. Every list tool supports filtering, sorting, and search, and each response states which filters actually ran, so agents reason over precise data instead of a page scrape.
Why we built it
Revenue teams are wiring AI agents into their daily work. Claude drafts the pipeline review, an Agent SDK workflow does account research overnight, Claude Code pulls numbers while an operator builds a report. Anthropic's Model Context Protocol became the standard way to give those agents tools (Anthropic), and our customers asked the obvious question: why can my agent read our CRM and our docs, but not our ABM platform?
Until now the answer everywhere in the ABM category was exports. Download a CSV of revealed companies, paste campaign stats into the chat, screenshot a dashboard. That workflow is slow, stale by the time it is read, and useless for autonomous agents that run while you sleep.
The Abmatic AI MCP server closes that gap. Your agent asks a question in plain English; the server answers from the same live data your Abmatic AI platform dashboard reads. If you are evaluating platforms for an agent-driven GTM stack, this is the difference between a tool your agents can use and one they can only read about. See how it fits the broader stack in our guide to API-native ABM platforms and the composable stack.
The platform behind the tools
What makes the MCP surface useful is what sits behind it. Abmatic AI is the most comprehensive AI-native ABM platform: web personalization, A/B testing, account-level deanonymization and contact-level deanonymization, first-party intent scoring, Agentic Workflows, Agentic Outbound, Agentic Chat, ABM advertising, and bi-directional Salesforce and HubSpot sync all run in one system on a shared identity graph. When your agent asks a question over MCP, it is reading the same live objects those modules produce, not a warehouse copy that syncs nightly.
Skip the manual work
Abmatic AI runs targets, sequences, ads, meetings, and attribution autonomously. One platform replaces 9 tools.
See the demo →What your agents can do with it
- Morning pipeline brief. "List the accounts whose engagement score jumped this week, pull their recent visitor sessions, and draft a brief for my 9am." One prompt, live data, no tabs.
- Pre-call research. Before a discovery call, an agent pulls the account record, recent page views, campaign touches, and any meetings already booked, and produces a one-pager. Pairs well with the workflow in our AI agent account research playbook.
- Reveal digest to Slack. A scheduled Claude Code job calls
list_revealed_companiesevery morning and posts the companies that visited yesterday, ranked by fit score, into your sales channel. - Campaign reporting on demand. "How did the Q3 personalization campaign perform against the enterprise segment?" The agent calls
get_campaign_analyticsand answers with real numbers instead of a guess. - CRM sync monitoring. An ops agent checks
get_crm_sync_statuson a schedule and flags error spikes before anyone notices missing records.
Teams already using AI for account-based motions will recognize the pattern: the agent stops being a writing assistant and starts being an analyst. Our roundup of the best AI agents for ABM covers where that trend is heading. See it on your own data.
The full tool reference
Tools ship in two tiers. Tier 1 is the core surface; Tier 2 is the comprehensive wave. Every tool is a read, scoped to the single account your API key belongs to.
| Tool | Tier | What it does |
|---|---|---|
whoami | 1 | Confirms the key works; returns account id, name, website, and plan tier. |
list_accounts | 1 | Lists and filters the companies Abmatic AI tracks, with fit, engagement, and overall scores. |
get_account | 1 | Looks up one company by website or id. |
list_contacts | 1 | Lists people, optionally scoped to one company or filtered by name or email substring. |
list_campaigns | 1 | Lists campaigns, optionally filtered by status. |
get_campaign | 1 | One campaign's configuration and summary. |
get_campaign_analytics | 1 | Performance metrics for one campaign. |
list_revealed_companies | 1 | Companies recently de-anonymized from your website traffic. |
get_analytics_summary | 1 | Account-level rollup: visitors, reveals, engagement. |
list_visitors | 2 | Raw visitor sessions (individual website visits). |
get_campaign_sequence | 2 | SDR sequence steps and tasks for a campaign. |
list_conversions | 2 | Conversions, list and detail. |
get_conversion_analytics | 2 | Conversion performance rollup. |
list_target_groups | 2 | Target groups and audiences. |
list_opportunities | 2 | Opportunities, list and detail. |
get_engagement | 2 | Engagement signals for your account. |
list_reports | 2 | Saved reports, list and detail. |
list_dashboards | 2 | Dashboards and their widgets. |
get_insights | 2 | Account insights. |
list_labels / list_tags | 2 | Labels and tags defined in your account. |
get_website_context | 2 | Your site knowledge and website context. |
list_meetings | 2 | Meetings booked. |
get_crm_sync_status | 2 | CRM sync status only: last-sync time, record counts, error summaries. Never returns credentials. |
Every list tool accepts sort_by, order_by, and fields, plus q search, score bounds, and time windows, and returns an applied block stating which filters and sort actually ran. That last detail matters for agents: they can verify the query they asked for is the query that executed. Book a demo to see the full surface live.
The security model
Handing an autonomous agent access to GTM data is a real decision, so the server is conservative by design:
- Read-only, no exceptions. Every tool is a GET. There is no write, update, or delete surface at all.
- Account-scoped keys. A key reads exactly one Abmatic AI account. Tenant isolation is enforced on every request.
- Treat the key like a password. It grants read access to the account's exposed data regardless of per-user dashboard roles. Store it in a secret manager, rotate it from the dashboard if it may be exposed; revocation is immediate.
- Rate limiting and audit logging. 100 requests per minute per key, with clean retry hints. Every request logs the account, key prefix, endpoint, and timestamp, never the full key and never PII values.
- Sensitive data is structurally excluded. User and role records, integration OAuth tokens, enrichment caches, and internal plumbing are not reachable through any tool. CRM integration data is exposed as sync status only.
How to connect
Setup takes about five minutes: mint an API key from the Abmatic AI dashboard, add https://mcp.abmatic.ai/mcp as a connector in claude.ai or Claude Code, and ask your first question. The full walkthrough for all three client types, including troubleshooting, is in our step-by-step setup guide, and the MCP server page has the short version.
API keys are available on paid Abmatic AI plans. If you are not a customer yet, book a demo: we will connect Claude to a live account during the call and run your questions against real data.
FAQ
What is the Abmatic AI MCP server?
A remote Model Context Protocol server at https://mcp.abmatic.ai/mcp that gives AI agents read-only access to your Abmatic AI account data: accounts, contacts, campaigns, de-anonymized visitors, conversions, opportunities, and analytics, through 28 tools over Streamable HTTP.
Which AI assistants and agents can connect to it?
The claude.ai custom connector, Claude Code, and agents built on the Claude Agent SDK or Managed Agents all connect today. Any MCP client that supports the Streamable HTTP transport and bearer authentication can use the server.
Can an agent modify or delete data through the MCP server?
No. All 28 tools are reads. There are no write, update, or delete tools, so a connected agent can query data but can never change campaigns, contacts, settings, or CRM sync.
How do I authenticate?
With an account-scoped Abmatic AI API key sent as a bearer token. You mint the key from the dashboard under Account, API Key; it is shown once at creation, and you can rotate or revoke it at any time with immediate effect.
Is the MCP server available on every plan?
API keys, and therefore MCP access, are available on paid Abmatic AI plans. Book a demo to see the server running against live data before you buy.
What data is never exposed through MCP?
User and role records, admin surfaces, integration OAuth and refresh tokens, third-party enrichment caches, global settings, and internal AI plumbing are structurally excluded. CRM integration data is exposed as sync status only, never as credentials.
Does the server support filtering and search?
Yes. Every list tool accepts sorting, field selection, text search, score bounds, and time windows, and returns an applied block stating exactly which filters and sort ran, so agents can trust the result set.
Ready to put your ABM data to work in every agent? Book a demo and connect Claude to your account this week.
