Model Context Protocol

Your CRM, in your AI workflow

Ask your CRM.
Work from the answer.

Find a lead, read the conversation, prepare the next step. Connect your AI client to 36 explicit tools in your own Outbound OS installation.

Connect your AI client Browse the tools

ChatGPT · Codex · Claude · compatible remote MCP clients

Outbound OS MCPExample · sample records

You ask

Which contacted leads have been quiet the longest?

Read toollist_leads

{
  "status": "contacted",
  "sortBy": "lastMessageAt",
  "sortOrder": "asc",
  "limit": 3
}

Returned records

Meridian DistributorsContacted · 9 days ago
Adler HandelContacted · 8 days ago
Northwind SupplyContacted · 6 days ago

Read the records. No message sent.

From context to next action

Three useful ways to put it to work.

01 / Read

Get back into a conversation.

Find the lead, read message history and inspect the timeline before writing a reply.

list_leadslist_messageslead_timeline

02 / Prepare

Build a campaign to review.

Inspect the audience and create a draft. Starting the campaign is a separate action with real consequences.

campaign_preview_leadscreate_campaign

03 / Inspect

Check the state of the system.

Read sender readiness and sending state. Investigate reported activity before deciding what to change.

whatsapp_accountssystem_status

Know what you connect

Reading context and sending a message are different permissions.

The client proposes tools. Your installation controls access. Your client’s approval settings decide when it asks you to continue.

Read the security model

A read returns CRM data.

Leads, messages and analytics can become context for the model you use. Connect a provider you trust with that information.

A write changes the workspace.

Send tools and campaign controls are live operations. Review recipients, content and sender before approving. Provider acceptance does not prove delivery.

Provider rules still apply.

The WhatsApp service window, approved templates, opt-outs and account limits apply to tool calls just as they do in the dashboard.

Make the connection

Start with one read. Verify the whole path.

The optional MCP container connects to your CRM API. Give it an HTTPS endpoint, configure authentication, then add that endpoint in your AI client.

Server & client setup guide

01
Run the MCP service.
Start the Docker MCP profile with separate service and client credentials.
02
Add your server to the client.
Use your own HTTPS URL ending in /mcp. Authenticate with the mode your administrator configured.
03
Run a read-only request.
A returned tool result verifies more than a successful health check. Do this before approving any write.

Your first read-only request

Use Outbound OS to list the first 3 contacted leads, sorted by lastMessageAt ascending. Show their names, status and last activity. Read only; do not send messages or change any records.

The server URL is not a credential. Keep tokens in the client’s credential configuration, never in a prompt.

Tool reference

All 36 tools, by job.

Exact names exposed by the server. Read and write tools share this catalogue.

Browse tools

Leads12

list_leads
get_lead
create_lead
update_lead
delete_lead
import_leads
list_tags
lead_timeline
lead_assign
lead_set_status
lead_toggle_pause
bulk_lead_action

Campaigns9

list_campaigns
get_campaign
create_campaign
update_campaign
delete_campaign
start_campaign
pause_campaign
campaign_preview_leads
campaign_ab_results

Messages5

list_messages
send_whatsapp
send_whatsapp_file
list_media_files
send_email

Analytics5

stats_overview
analytics_funnel
analytics_campaign_roi
analytics_email_performance
whatsapp_costs

System5

system_status
system_pause_resume
whatsapp_accounts
whatsapp_templates
list_users

The website’s tool catalogue is checked against the MCP server in CI.

FAQ

Connecting your AI client

What does the MCP connection add?

It gives a compatible AI client tools that read or change records in your Outbound OS installation. The assistant requests a tool, the server calls the CRM API, and the result returns to your client. It is not a separate copy of the CRM.

Can I connect ChatGPT, Codex or Claude?

Use a client that supports a remote Streamable HTTP MCP server and your configured authentication mode. Client editions and workspace policies can restrict custom servers. The setup guide explains the server endpoint, bearer-token connection and OAuth flow; check your client’s current requirements.

Can an AI client send messages?

The server includes send and campaign-control tools, which can change real data or send through your connected accounts. Authentication is not approval for an individual send. Configure your client to ask before write actions and review the recipient, sender and content.

Is my dashboard role automatically the MCP role?

Do not assume it is. The service-credential connection uses the server’s configured CRM access, rather than inheriting your dashboard session. Your administrator must choose and secure the authentication mode. Treat a shared service credential as privileged.

What reaches the AI provider?

Tool results are returned to the AI client and can become context for its model. Connect only a client and provider you trust with the data it requests. Keep service tokens out of prompts, public pages and source control.

Keep the CRM where you run it

Your workspace. Another way to work.

Run Outbound OS on your infrastructure, connect the channels you need, and add the MCP service when you are ready.

Run Outbound OS

Ask your CRM.Work from the answer.