Guide
How to give Claude read and write access to your tools
Read access lets Claude look things up. Write access lets Claude get work done. This guide explains how both work through a custom MCP connector, how auth keeps that access scoped to what you already have, and what stays gated behind your approval.
Key takeaways
- Read access lets Claude pull data; write access lets it create and update records. A custom connector exposes both as separate tools.
- The connector authenticates as your account, so Claude inherits your existing permissions and never gets more access than you already have.
- Create and update flow inside a conversation you start. Deleting records and sending messages stay behind a confirmation you approve.
- Built-in directory connectors are often read-leaning. A custom connector covers the full read-write surface, including internal tools.
- You revoke access the same way you revoke any account: disable the login or rotate the credential.
How do I give Claude read and write access to my tools?
Giving Claude read and write access to your tools means connecting each tool through a custom MCP connector that exposes both read tools and write tools. Read tools pull records and data. Write tools create and update records. The connector authenticates as your account, so Claude acts inside the permissions you already hold.
Most tools expose an API. A connector wraps that API in the Model Context Protocol so Claude can call it during a conversation. For tools that are not in Claude's connector directory, or for internal and custom systems, the connector has to be built for your exact setup. That is the work we do for you, so you do not run or maintain any server.
What is the difference between read access and write access?
The difference between read access and write access is direction. Read access lets Claude look at your data: search records, pull a report, summarize a pipeline. Write access lets Claude change your data: create a contact, update a deal stage, log an activity. Most real work needs both, which is why a connector exposes them as distinct tools.
| Read access | Write access | |
|---|---|---|
| What Claude does | Looks up and pulls data | Creates and updates records |
| Example actions | Search contacts, summarize a report, list open deals | Add a contact, change a deal stage, log a note |
| Reversibility | Nothing changes in your system | Create and update are recoverable; deletes are gated |
| Confirmation | Runs freely inside your conversation | Create and update flow; destructive actions need approval |
Is it safe to give Claude write access?
Giving Claude write access is safe when that access is scoped. The connector authenticates as your account and inherits your existing permissions, so Claude can never reach data or actions you cannot reach yourself. Create and update flow inside a conversation you start, while destructive and outbound actions stay behind a confirmation you control.
Safety comes from three layers working together: the connector never holds more access than the account it runs as, write tools are separated from read tools so permissions stay legible, and the riskiest actions are held behind a human-in-the-loop step. Here is how those layers map to what Claude can and cannot do on its own.
- Flows freely: reading records, creating a record, updating a field. These are the everyday actions that make Claude useful inside your tools.
- Gated behind your approval: deleting a record, sending an email, posting a message. Claude proposes the action and waits for you to confirm before it runs.
- Never possible: anything outside your account's permissions. The connector cannot grant access you do not already have.
How does authentication keep that access scoped?
Authentication keeps access scoped by connecting as your account, using OAuth where your tool supports it or a scoped API key or token where it does not. The connector inherits your existing roles and permissions, so Claude sees exactly what you see. When you change a role or revoke a credential, the connector's access changes with it.
This is also why offboarding stays simple. Because the connector rides on real auth, you do not manage a separate access list for Claude. Disable the account or rotate the token and the connector stops working, the same way any other integration would. For a deeper walkthrough of the build and the security model, see the complete guide to custom Claude connectors.
How do I set up read and write access for a specific tool?
You set up read and write access by scoping the tool, building a custom MCP connector for it, and adding that connector in Claude. We map the records Claude should read and the actions it should take, build and host the server with separated read and write tools, then maintain it as your tool's API and schema change.
The same pattern works whether the tool is a popular SaaS app or an in-house system. If you want to see it applied end to end, the custom CRM connector walks through mapping objects, scoping write actions, and reaching a system that lives behind a firewall.
Frequently asked questions
How do I give Claude read and write access to my tools?
Is it safe to give Claude write access to my systems?
Can Claude delete records or send messages on my behalf?
Does Claude's built-in connector already have write access?
What happens to access when an employee leaves?
Tell us what Claude should be able to do.
Book a 30-minute scoping call. We'll map the records Claude should read, the actions it should take, and what stays gated behind your approval.