Sightpane
Try the demoBook a call

For clients

For agencies

AI

Every feature, on one page.

See the product
Book a callTry the demo

Guide

How can agencies use AI on client projects safely?

Let AI read the project and draft, and keep a person in charge of anything a client sees and anything that changes. In practice that means answers with sources, drafts a person publishes, a confirm step before every change, and agents connected through scoped, logged access.

Updated

Where does AI help an agency with client work?

Most of the gain is in the routine work around a project, not the craft itself.

  • Answering status questions from the project's own data, at any hour.
  • Drafting the weekly update and replies to client messages.
  • Sorting the requests queue, and finding what is overdue or missing across projects.
  • Filing a client's change request in the right place, in their own words.
  • Pulling the decisions and next steps out of call notes.

What are the risks of AI on client projects?

Four come up again and again. An answer that sounds right but is not in the data. A change made without anyone asking for it. A model that can see another client's project. And no record of what an agent did, or when.

Each one has a plain fix, and the fixes are the same whether the AI is built into your tools or an agent your team connects.

What rules keep AI safe on client work?

  1. Answer from the data, and show the sources

    Every answer should cite the update, message or approval it came from, so anyone can check it.

  2. Confirm before every change

    The AI proposes, a person confirms. Nothing is filed, moved or sent on the model's own say.

  3. Draft, do not post

    Updates and replies written by AI wait in drafts until a person on the team publishes them.

  4. Scope access to the person

    The AI sees only the projects the person in front of it can see, and checks that again on every action.

  5. Give each agent its own token

    Each connected agent gets its own access level and its own list of projects, and can be revoked on its own.

  6. Log every call

    Keep a record of what each agent read and changed, with the time and the project.

What is MCP, and why does it matter for agencies?

MCP, the Model Context Protocol, is an open standard for connecting AI agents to the tools where work lives. An agency that already uses an agent such as Claude Code can point it at its client platform, and the agent can read projects, file requests and draft updates there instead of in a copy of the data.

In Sightpane the MCP endpoint is built in. Each agent gets a token with an access level. Read lets it list and read. Draft also lets it file requests and write drafts. Publish, which only an owner can grant, lets it post to a client directly. A token can be limited to chosen projects, and every call is logged on the Integrations page.

How do you connect an AI agent to Sightpane?

  1. Create a token

    In Settings, open Integrations and create a token for the agent. It is shown once.

  2. Choose its access and projects

    Start with Read, and limit it to the projects it needs.

  3. Add it to your agent

    The Integrations page shows the exact command for Claude Code and the settings for Claude Desktop, with your endpoint filled in.

  4. Review its drafts

    Read what the agent drafts before it reaches a client, and check the call log now and then.

Questions

No. Every change it proposes shows as a card the person confirms or cancels, and nothing changes until they confirm.

No. Sightpane AI sees only the projects the person can see, and every action checks that access again.

Sightpane AI is built on Claude, by Anthropic.

Only with Publish access, which only an owner can grant. Without it, an agent's updates and messages wait in Drafts for a person.

See it on a real project.

Open the live demo with example projects, or book a short demo and we set Sightpane up with you.

Book a callTry the demo