AutomationApril 9, 20269 min read

We Built a Custom MCP Server for GHL.
Now Claude Controls the CRM.

30+ tools. Full CRM, conversation, pipeline, calendar, invoice, and workflow access — all callable from a single Claude session. This is what it looks like when the AI is not just answering questions but actually operating the system.

What a Custom MCP Server Actually Is

Model Context Protocol is Anthropic's standard for giving Claude access to external tools. Most implementations connect Claude to a generic API or a documentation search. What we built is different. The Codexo GHL MCP server is a bespoke connector between Claude Desktop and a live GoHighLevel account — built specifically around the operations of service businesses.

The server exposes 30+ tools across every major GHL object: contacts, conversations, pipelines, calendars, invoices, workflows, and sub-accounts. Claude can call any of them in sequence, in a single session, without switching tabs or copying data between systems. You describe the task. Claude executes it against the actual CRM.

The difference between a CRM with an API and a CRM that Claude operates is not technical. It is operational. One is a database. The other is a system that acts.

What 30+ Tools Actually Means in Practice

Here is a concrete example. A service business operator opens Claude Desktop in the morning. They type: “Pull all contacts who came in from the pool service form in the last 7 days, check which ones have not had a follow-up call logged, and create a task for each one.”

Without the MCP server, that is a 15-minute GHL task involving contact filters, pipeline views, conversation checks, and manual task creation. With the MCP server, Claude executes it in under 60 seconds and reports back a list of contacts with tasks created. No switching tabs. No copy-paste. No missed records.

The tool coverage is what makes this useful. Partial access means partial utility. If Claude can read contacts but cannot create tasks, the workflow still requires manual steps. If Claude can see the pipeline but cannot move stages, you are back to doing it yourself. 30+ tools means the common operational tasks are fully executable without a human in the loop.

The tool categories we built out: contact search, create, update, tag, and merge. Conversation read and send. Pipeline stage moves and opportunity creation. Calendar availability check and appointment booking. Invoice creation and payment links. Workflow enrollment and removal. Sub-account management for agency use.

Why This Changes the Operator Relationship With the CRM

Most service business operators know their CRM is underused. They know the pipeline is not accurate. They know there are contacts sitting in stages they should have left 60 days ago. They know follow-up tasks are not being completed. The problem is not awareness — it is time. Cleaning and operating a CRM manually is work that gets deprioritized when there are actual jobs to run.

The MCP server changes the cost of CRM operations. When Claude can do the audit, the cleanup, and the bulk updates in a single conversation, operators actually do it. The friction is gone. The result is a CRM that reflects reality — which is the entire point of having a CRM in the first place.

For the agencies and consultants we build this for, the MCP server also changes what is possible on behalf of clients. A GHL agency managing 20 sub-accounts can run cross-account audits, bulk pipeline health checks, and systematic contact enrichment tasks that would be manually impossible at that scale. Claude operates the system. The operator directs the strategy.

We are not predicting that AI will replace CRM workflows. We are watching it happen in the specific, concrete way that matters: operators spending less time inside the tool and more time making decisions about what the tool should do next.

See the MCP server operate a live GHL account.

Thirty minutes. We will demo the full tool coverage against a real account and show you what your operators stop doing manually.

Book a Strategy Call