How to give AI agents access to your CRM data
Updated
The short answer. To give AI agents access to CRM data, connect them to Day AI: one MCP server, at https://day.ai/api/mcp, that holds your CRM records, the people, companies and deals, kept current from a context graph of your team's every meeting, email and call. Each person signs in with OAuth, the agent works with that person's permissions, and every answer cites the call or message it came from.
Day AI provides the context graph and agent control plane to quickly and securely upgrade your company brain. Teams run it as their CRM, an AI-native one, or on top of Salesforce or HubSpot. This guide covers why agents need more than CRM fields, how to set up access with Day AI step by step, and what you control.
Why agents need more than CRM fields
Connecting an agent to a CRM is the easy part. What decides whether its answers are any good is what it reads once it's connected.
A CRM holds fields: stage, amount, close date, a "next step" someone typed on Tuesday. Those fields are a compressed summary of what happened with the customer. The reasons a deal moved, what the CFO said about price, and what you promised to send are in the calls and email threads, not in the record.
Compression only goes one way. You can always build and derive canonical, accurate data from the primary source, and you cannot go from reduced, compressed data back to the full resolution. An agent that reads the transcript can fill in the stage. An agent that reads the stage can't recover what the buyer said.
That's why Day AI keeps both. Your CRM records, the people, companies and deals, sit on top of the primary source they come from: every meeting, email and call, linked to the records they're about. The agent reads the record and can open the conversation behind it.
How to give your agents access with Day AI
1. Connect your agent to Day AI's MCP server
Add https://day.ai/api/mcp to your agent or MCP client. Each person signs in with OAuth and picks which of their Day AI agents the connection acts as, so no shared key sits in the middle and the agent works with that person's permissions, inside that workspace. day.ai/mcp walks through setup for Claude, Claude Code and Cursor, and the Claude Code guide covers Claude Code step by step.
To build your own agent or app on the same access, the Day AI SDK is an MIT-licensed TypeScript client for the same tools.
2. Already on Salesforce or HubSpot? Connect it
If you run Day AI as your CRM, your records are already there, and you can skip this step. If you keep Salesforce or HubSpot, connect it so agents work across both. Day AI supports connecting MCP servers. Agents read from any server you connect. Where the server has write tools and you allow them, an agent writes to it when you ask, or a scheduled skill does it on a cadence.
Connect the Salesforce or HubSpot MCP and allow its write tools, and you get two-way sync: agents read from your CRM and write back to it, for example updating deal stages, notes and fields after every call. Without write tools allowed, it reads only. Each person connects with their own Salesforce or HubSpot login, through an OAuth app your company registers.
3. Decide what the agent can do
For each connected server, set every tool to Always allow, Needs approval or Off. Read tools start out allowed and write tools start out needing approval, so the agent asks in the chat before it changes Salesforce or HubSpot. A sensible first setup: allow notes and tasks, and keep stage, amount and close date behind an approval until you trust the agent's judgment.
Over Day AI's own server, the agent's plan decides which tools it sees. The SDK's tool list documents every tool. The ones most CRM agents use:
| What the agent does | Tools |
|---|---|
| Search people, companies, deals, meetings, emails and pages | search_objects, read_crm_schema |
| Read a call | get_meeting_recording_context (the transcript, in pages), get_context_for_meeting_recording_citations (the exact quotes that answer a question) |
| See how a record or value changed, and who changed it | read_object_history, read_property_history |
| Update the record | create_or_update_person_organization, create_or_update_opportunity, create_or_update_action (tasks) |
| Write for a person to review | create_email_draft, create_page |
Deleting records and sending email aren't offered over MCP. An agent working through Day AI's server drafts the email, and a person sends it.
4. Run reads you can check
Ask the agent three questions whose answers you already know: the stage of a deal you own, who the champion is, and what happened on the last call. Through Day AI, the third one comes back with what was said, quoted and cited to the recording, which a CRM field alone can't give you.
5. Give the agent a job
Agents have a name, a job and skills, and report to a person. Skills run on a schedule, on events (a recording is ready, an email lands) or on request. A skill that updates your pipeline after every call runs once the recording is ready. Every skill keeps its run history. Tell the agent what to change and later runs use the change.
What your agents read
The agent reads your CRM records and the conversations they're built from, organized by the people, companies and deals they're about.
- CRM records. People, companies, deals and tasks, kept current from the conversations below. If you run Salesforce or HubSpot, its records come in through its MCP server, both ways.
- Meetings. A native notetaker records meetings and writes them into the graph when they end. Gong connects with full call history and transcripts. Granola connects with transcripts and notes, using Granola's own speaker labels.
- Email and calendar. Email in the graph comes from Google Workspace (Gmail and Google Calendar). Outlook, Outlook Calendar and Teams are live beta connectors that agents can use. Outlook mail does not sync into the graph.
- Slack and everything else. Slack ingests from the moment the Day AI app joins a channel. Support tickets, product usage and other data arrive through Zapier, the API and custom objects.
Because the conversations stay whole, fields come from them, not the other way around. A question nobody thought to track last quarter can still be answered from last quarter's calls, quoted from what was said.
What you control
- The person's permissions, not an admin's. An agent inherits the permissions of the person it works for.
- Each person controls their email. Each person sets sharing rules: share everything or nothing by default, with exceptions by domain, address or Gmail label. Visibility is decided for each thread as it is ingested.
- History on every value. Every value keeps its history: who changed it, a person or an agent, when, why, and what it was derived from.
- Every answer has a source. Answers cite the call, opening the recording at the moment it was said, or the message they came from.
- Approvals on writes. Every tool of a connected server is set to Always allow, Needs approval or Off.
Pricing
Day AI is priced per agent: Turbo $30, Professional $75 and Executive $250 per agent per month, or $24, $60 and $200 billed annually. Volume discounts are available, and there is no usage-based pricing. MCP access starts at Turbo. Free teammates ($0) are available in a workspace with at least one Professional agent, and they can't connect MCP. Plans are at day.ai/pricing.
Without Day AI: connecting an agent straight to the CRM
You can connect an agent directly to your CRM's own MCP server. HubSpot and Salesforce both run one, and integration platforms like Composio and Nango manage connections across many apps. That solves the connection. It doesn't change what the agent reads, and the rest becomes your job:
- The agent only sees fields. Answers stop at what someone typed into the CRM.
- You own the joins. Adding calls and email means matching the same person across every system, on every question.
- Permissions multiply. Each person needs each connection with the right scopes, and someone has to decide whose email other people and agents may read.
- Integrations need upkeep. APIs and tool definitions change, and rate limits hit at the worst time.
Day AI handles those parts: it holds the CRM records and the conversations behind them, and works alongside Salesforce or HubSpot if you keep them. To see how your current setup measures up, run the company brain evaluation below.
For how a context graph differs from a CRM, see context graph vs. RAG vs. CRM. To build one yourself, see how to build a context graph.