How to give every agent on your team the same company context
Each person's agent works from a different, partial picture of the company. Here is how to give all of them one shared source of facts.
To give every agent on your team the same company context, put that context in one shared place and connect every agent to it, instead of loading each agent separately. MCP, the open standard for connecting agents to tools, makes this possible across Claude, ChatGPT, Cursor, Codex and others.
The usual alternative is that each person builds their own pile of notes, files and prompts. It works for one person and falls apart for a team.
Why agents on the same team disagree
Picture three teammates, each with a capable agent. One uses Claude, one ChatGPT, one Cursor. Ask all three about the same customer and you may get three different answers. Not because the agents differ, but because each was fed different material:
- One teammate pasted in last week's call. Another did not.
- One has a project full of old documents. Another has nothing.
- Nobody's agent has seen the call a fourth colleague took yesterday.
The fix is not a better prompt. It is a single source they all draw from. For the background, see what is shared memory for AI agents.
The approach, in five steps
This works whatever tools you pick.
- Decide what counts as company context. For most teams: what was said in meetings, what is in email threads with customers and partners, and who is meeting whom. Start with those three.
- Pick one place for it. One store that agents query, not a folder per person. If it needs hand-curation, it will go stale, so prefer one that fills itself from the work that already happens.
- Capture automatically. Meetings, email and calendar already exist. Connect them rather than asking people to write summaries.
- Connect each agent to the same store over MCP. Each person keeps the agent they prefer. They all point at the same source.
- Set the access rules before you invite everyone. Shared does not mean everyone sees everything.
The rules that matter
Facts should carry their source
If every agent answers from the same store, one wrong entry is wrong everywhere. Insist that facts keep a quote, a date and a link to where they came from, so anyone can check. See saved isn't searchable for how to test this.
Scope access per person
A teammate's agent should see what that teammate can see, no more. A shared brain that shows everyone everything is a liability, especially for private email.
Let agents write, not just read
Agents learn things in conversations: a corrected job title, a new contact, a decision. If they can only read, every one of those has to be re-entered by hand. Write access, signed in as the teammate, keeps the store current.
How to do it with Nysa
Nysa is built for exactly this. It turns a team's meetings, email and calendar into one shared brain that every agent works from:
- Capture. A notetaker joins the calls you choose as a visible participant and transcribes them. Gmail and Google Calendar sync in. Nysa supports Google accounts today.
- Enrich. Nysa turns this into facts with sources: quotes, dates and where they came from.
- Retrieve. Every agent reads the same brain over MCP. Agents can also write back, each signed in as its teammate.
Each agent sees only what its teammate can see, and raw email stays private to its owner. Nysa works with Claude, ChatGPT, Codex, Cursor, OpenClaw, Hermes, and any agent that connects over MCP. Onboarding a whole team takes under 5 minutes.
To connect each person's agent, follow the page for their tool:
- Claude: Connect Claude. Find Nysa under Customize, then Connectors, select Connect, and sign in with Google.
- ChatGPT: Connect ChatGPT. Open Apps, search for Nysa, select Connect, and sign in with Google.
- Claude Code:
claude mcp add --transport http nysa https://mcp.usenysa.com/mcp - Codex:
codex mcp add nysa --url https://mcp.usenysa.com/mcp
Every connection points at the same MCP address, https://mcp.usenysa.com/mcp. That is the whole idea: many agents, one source.
Checking that it worked
Pick one customer and ask each teammate's agent the same question: "What do we know about this account, and what is the source?" The answers should match, and each should cite the same calls or emails. If they differ, one agent is not connected to the shared source, or is seeing different material because of different permissions. Both are easy to diagnose.
For the Claude-specific walkthrough, see how to give Claude your team's meeting context. For ChatGPT, see how to connect ChatGPT to your company's meetings, email and calendar.
FAQ
Do all my teammates have to use the same AI agent?
No. Any agent that connects over MCP can use the same source. One person can use Claude, another ChatGPT, another Cursor, and they all see the same facts, within each person's permissions.
If every agent shares one brain, can a junior teammate's agent see executive email?
Not with Nysa. Each agent sees only what its teammate can see, and raw email stays private to its owner. The team gets the facts drawn from it, not the messages.
What if an agent writes something wrong?
That is why facts should carry sources. Because each fact points back to its quote, date and origin, you can check it and correct it rather than trust it blindly.