How do you connect Claude to your CRM?
To connect Claude to your CRM, you need a CRM that offers an MCP server. In Claude Code you add the server with one claude mcp add command and an API key. In Claude Desktop or claude.ai you add it as a custom connector and an admin approves it. After that, Claude calls the CRM's tools directly.
What do you need before you connect Claude to a CRM?
- A CRM with an MCP server and an https endpoint Claude can reach
- Claude Code, Claude Desktop or claude.ai
- Admin rights in the CRM for one key or one connector approval
- One concrete job you want Claude to do first
The last point sounds soft, but it's the one people skip. If you don't know which job Claude should do, you end up testing random prompts and the whole thing feels like a toy.
How do you connect Claude Code to a CRM?
Claude Code registers remote MCP servers with the claude mcp add command and the http transport, and you can pass headers such as an Authorization bearer token [1]. For Sigma CRM it looks like this:
claude mcp add sigma-crm --transport http https://YOUR-SIGMA-CRM-DOMAIN/api/mcp \ --header "Authorization: Bearer sigma_sk_..."
The key dialog in Sigma CRM shows this command ready to copy, so nobody has to type the key by hand. Then run /mcp in Claude Code to see that the server is connected and which tools it offers.
How do you connect the Claude app as a custom connector?
Claude Desktop and claude.ai connect to remote MCP servers as custom connectors: you open the connector settings, add a custom connector and paste the server URL [2]. The server has to be reachable over https from the internet for that to work.
In Sigma CRM no key is pasted in this flow. Sigma CRM opens, a workspace admin picks the workspace and the scopes and presses Allow, and the connection can be removed again at once.
Which prompts should you try first with Claude in your CRM?
| Prompt | What Claude does | Risk |
|---|---|---|
| Research acme.example and log buying signals | Finds or creates the account, saves research, logs signals with a source | Low, writes only new signals |
| Review my open deals without a next step | Lists deals and recent activity, proposes next steps | Read only |
| Summarise this week's pipeline as a note | Reads deals and activities, saves one note | Low |
| Import this CSV, dedupe on email, show me first | Runs a dry run, then imports after your approval | Medium, needs confirmation |
The full list with the real tool names is on the MCP server page, and the Claude CRM page shows the three jobs teams ask for first.
How do you keep Claude's access to the CRM safe?
OWASP describes excessive agency as the risk that a model can do damage because it has too much functionality, too many permissions or too much autonomy [3]. So I'd start with the smallest scope that does the job, which is often a read key, and only add write access once you trust the routine.
Bulk actions are the other thing to watch. A good CRM MCP server asks for confirmation before merges, imports and sequence enrollments, and it tags every write so you can see later what came from Claude.
What does a Claude connection look like in practice?
It looks like one key and one routine, not a big project. At JadenX we built the Sigma CRM MCP server so that the API key belongs to one workspace and not to a person, reaches only shared lists and segments and fails closed. Writes are idempotent, so a retried call does not create duplicates, and every write shows up in the record history as an MCP write. The same server runs in our cloud and on-premise, so the steps above are the same in both.
Frequently asked questions
Can Claude connect to any CRM?
Only to CRMs that offer an MCP server, or through an MCP server someone builds on top of the CRM's API. Sigma CRM has one built in.
Do I need Claude Code, or is the Claude app enough?
Both work. Claude Code uses an API key and suits routines and developers. The Claude app uses a custom connector with an admin approval and suits everyday questions.
Does an on-premise CRM work with the Claude app?
Only if the endpoint is reachable over https. Claude Code just needs network access to the endpoint, which also works inside a company network.
Who pays for the model usage?
Claude usage is billed by Anthropic under your own account. CRM licensing is separate, see Sigma CRM pricing.
References
- Anthropic, "Connect Claude Code to tools via MCP", Claude Code Docs, 2025. https://code.claude.com/docs/en/mcp
- Anthropic, "Getting started with custom connectors using remote MCP", Claude Help Center, 2025. https://support.claude.com/en/articles/11175166-getting-started-with-custom-connectors-using-remote-mcp
- OWASP GenAI Security Project, "LLM06:2025 Excessive Agency", OWASP, 2025. https://genai.owasp.org/llmrisk/llm062025-excessive-agency/
About this article: I work with a set of AI agents that research sources and turn my notes and the Sigma CRM documentation into a draft. The product facts come from our own documentation and code. The opinions are mine.
If you want help connecting Claude to your CRM and picking the first routine, talk to me at JadenX.