Tools
Your agent’s instructions and knowledge decide what it knows. Its tools decide what it can do: look up today’s news, read the page a client sends, make a slide deck, check a booking in your CRM, or add a new lead to your email list.
How tools work
Think of an agent as a new team member. You’ve told them who they are and what they know. Tools are what you let them do on the job.
There are two kinds:
- Built-in tools are skills MindPal gives every agent, such as searching the web or making images. There’s nothing to set up: you switch them on.
- Your apps and services are the accounts your business already runs on, such as Gmail, Notion, Zapier, Make or your own systems. You connect each one once, and then choose what each agent may do with it.
The connection and the permission live in different places:
- The connection belongs to your workspace. You connect Gmail once, like handing over a key.
- What each agent may do belongs to the agent. A support agent might only search your emails, while a sales agent may also write drafts. Both use the same Gmail connection.
Your agent decides on its own when to use a tool. It reads what each tool does and your instructions, then picks. Like a person, it chooses better from a short, clear list than from hundreds of options. Giving each agent only the tools its job needs is the single biggest thing you can do to make tools work well.
Start with built-in tools
Most agents only need built-in tools. Open your agent, go to Tools, and switch on what its job needs:
| If you want your agent to… | Switch on |
|---|---|
| Answer with up-to-date facts, news or local places | Web search |
| Read a web page, a YouTube video or a feed someone shares | Read websites |
| Make images | Image generation |
| Write a document the reader can edit | Text document |
| Make a web page, a mind map or flashcards | HTML page |
| Make a slide deck, or an Instagram-style carousel | Slides, Carousel |
| Work with numbers or analyse a spreadsheet | Code sandbox |
| Build a small web app (beta) | App builder |
| Work through a long, multi-step task without losing track | Task plan |
| Ask a visitor a quick question instead of guessing | Clarifying questions |
| Find and use your workspace’s other tools when it needs one | Find workspace tools |
New agents start with Web search, Read websites and Clarifying questions on. Switch off anything the agent doesn’t need.
A few of these only work in some places:
- Clarifying questions asks with choices to pick from on your published web chatbot.
- Find workspace tools doesn’t run on a published chatbot.
Connect your apps and services
Pick the right way in
Start from what you want your agent to do and what you already have:
| You want your agent to… | Connect it by |
|---|---|
| Work in an everyday app, like reading Gmail or finding a file in Google Drive | Signing in to the app. Needs a paid plan |
| Use automations you’ve already built in Zapier or Make | Zapier or Make, guided setup |
| Work with Notion, GoHighLevel, Apify, Tally, Firecrawl, DataForSEO or EmailListVerify | That service’s guided setup |
| Send information to a webhook, like a Zapier or Make “catch hook”, or call one web address | A link to call |
| Use an API your developer documented with an OpenAPI spec | An OpenAPI spec |
| Use a service that gave you an “MCP link” | An MCP server, then Another service |
Not sure? Search the catalog for what you want done, like “send email”, and MindPal shows the apps and services whose tools match.
Zapier and Make are the easiest way to reach an app MindPal doesn’t list. Build the action in Zapier or Make once, connect it to MindPal, and your agent can use it.
Connect it
Open the catalog
Go to Assets → Tools and click Add an app or service. You can also click Add in an agent’s Tools section, which connects it and puts it on that agent in one go.
Follow the short setup
For an app, you sign in to your account. Guided services either offer sign-in or open the right page on the service so you can copy the token or key they need. For your own link or API, you paste it.
If a service gives you an MCP link, choose An MCP server → Another service and paste its online Streamable HTTP link. MindPal asks you to sign in or paste a key when needed. Local servers that only run on your computer cannot be connected. MindPal uses the server’s tools, not its resources or prompts. Account sign-in requires the server to let MindPal register itself automatically.
Check what your agent gains
MindPal checks the connection straight away and shows the tools your agent will get, for example “Zapier gives your agent 3 tools”. If something’s wrong, it tells you what to fix in plain words, such as copying the link again.
Give it to your agents
Open each agent that should use it, go to Tools, and add it. Then press Update agent: tool changes, like any agent change, only take effect when you save.
Give each agent only what it needs
When you add an app or service to an agent, open it to choose exactly which of its tools that agent may use:
- Apps start with nothing switched on, because they often have hundreds of actions. MindPal suggests the ones most people use; pick what the job needs.
- Zapier, Make and other MCP services start with all their tools on, including ones you add later on the service’s side. For Zapier and Make, you usually choose the actions or scenarios in Zapier or Make itself.
- Your own APIs start with every tool on. Switch off the ones this agent doesn’t need.
Why it matters: the description of every tool you give an agent goes along with each reply. Thirty tools from one app is a lot to read, and MindPal marks a source like that with Many tools. Fewer tools mean faster, better choices, and less that can go wrong.
Teach the agent your own tools
When you add a link to call or an API, you’re writing the instructions the agent reads, so a few minutes here pays off.
- Say when to use it, in plain words. “Add a lead” is vague. “When a visitor shares their name and email and wants a follow-up call” tells the agent exactly when.
- Decide who fills in each piece of information. Let the agent fill in what changes each time, like the visitor’s name and email. Set what’s always the same, like
source: website chatbot, so the agent never has to think about it. Leave out what this tool shouldn’t send. - Keep keys out of the agent’s hands. A key or password is saved with the connection and sent only to that service. The agent never sees it.
- Try it before you rely on it. Try it sends a real request with sample values, so you see the answer before your agent ever uses it.
Your custom API tools made before October 30, 2026 were moved over automatically and keep working. You can now edit each one’s description and inputs the same way.
Tell your agent when to use its tools
Each tool comes with a description, but your agent’s system instructions are where you explain how tools fit its job:
Before answering a question about current events, use Web search.
When a visitor wants a follow-up call, ask for their name and email,
then add them with "Add a lead to the CRM". Confirm when it's done.
Never send an email without showing the visitor the draft first.You can type @ in the instructions to mention a tool by name. A tool you mention is added to the agent when you save. See Smart Resource Mentions.
Test before you share
Chat with your agent in MindPal before you publish it. Each time it uses a tool, the chat shows a step you can open to see exactly what it sent and what came back:
- “Searching the web…” while it works, then “Searched the web” when it’s done.
- ”· nothing found” when the tool worked but had nothing to return. Try a different request or source.
- “Couldn’t …” when the tool failed. Open the step to see why.
- “Stopped before finishing” when the reply ended while a tool was still running.
On a published chatbot, visitors see the same steps as one short, plain line each, like “Searched the web”, never your setup or a technical error. For a workflow form, you choose whether users see each tool’s details or one line per step. See Publish Workflows as Forms.
Keep your accounts and visitors safe
When an agent powers a published chatbot, anyone chatting with it can make it use its tools, with your connected accounts. If that agent can send email from your Gmail, so can a stranger on your website. MindPal reminds you with a Visitors use these tools note on the agent.
- Give public agents the smallest set of tools that does the job. Looking things up is safer than sending, changing or deleting.
- Ask for confirmation in the instructions before any action that sends or changes something.
- Use separate agents for your team’s internal work and for the public, so internal tools never reach visitors.
Connections are shared by the whole workspace, so changing or disconnecting one affects every agent that uses it. MindPal tells you which agents first.
When a connection stops working
Connections can stop working when a sign-in expires, a key is changed, or the service is down. When that happens, the agents that use it can’t use its tools until it’s fixed.
Go to Assets → Tools. Anything that needs attention is at the top, with what’s wrong in plain words and a Fix button. Usually the fix is signing in again or pasting a new key. If a service isn’t answering, try again in a moment, or check whether its link changed.
For more help, see Common Issues.
Plans and credits
- Apps you sign in to, like Gmail and Google Drive, need a paid plan. On the free plan they show a lock. The guided services and your own tools aren’t locked. See Pricing & Plans.
- Built-in tools may use MindPal credits, depending on what they do. See AI Credit.