MCP
Your company’s context, served to every AI.
Claude, Copilot, ChatGPT, Cursor — smart, but blind to what your company decided last Monday. In Parallel’s MCP server exposes your live organisational context to every agent, scoped to the person asking. One connection. No lock-in.
In short
In Parallel MCP is a Model Context Protocol server that gives Claude, Copilot, ChatGPT, Cursor and other AI tools your company’s live context: decisions, risks, to-dos and pages captured from meetings, each linked to its source. Every answer is scoped to what the person asking can already see. MCP access is included in every plan.
What it does
Every agent starts from reality, not a blank slate.
Current and cited, not a pile of files
Your AI reads the record your meetings keep current: decisions, risks, to-dos and pages, each with the date it was true and a link back to the meeting it came from. When an answer looks wrong, you can open its source and check.
Scoped to the person asking
Access is granted at four depths (status, storylines, observations, and the meetings and documents beneath), by project, named person or role. Your AI only sees what you can see, and an agent re-checks its grant on every call.
Platform-agnostic by design
MCP is an open protocol. Your world model is available wherever MCP is supported, today and as your stack changes. Vendors that make interoperability hard get dropped. We don’t make it hard.
How it works
One endpoint. Everything in context.
01 · Connect
Connect
One MCP endpoint per workspace. Add it to Claude, Copilot, ChatGPT, or Cursor the way you’d add any MCP server — minutes, not a project.
02 · Ask
Ask from any tool
"What did we decide about pricing?" works the same in every agent, answered from the same shared context, with sources.
03 · Stay current
Stay current automatically
The context updates as meetings happen and pages change. No exports, no syncing jobs, no context files to maintain.
Works with
Every agent platform that speaks MCP.
connection per workspace
Not one integration per tool per team. The endpoint is the integration.
agent platforms today
Claude, Copilot, ChatGPT, Cursor, Gemini — and whatever your stack adopts next, if it speaks MCP.
custom integrations to build
An open protocol instead of a vendor SDK. Leaving is as easy as arriving — which is why you won’t need to.
Before you ask
The questions every team asks first
- What is MCP?
- Model Context Protocol — the open standard AI tools use to read from external sources. In Parallel publishes your shared organisational context as an MCP server, so the AI tools your team already uses can draw on it directly.
- Is this safe to connect?
- The endpoint enforces the same permissions as the product: every answer is scoped to the person asking, and every claim carries its source. Nothing is exposed beyond what that user could already see in In Parallel.
- What goes into the context?
- Meetings and the threads around them. The notetaker extracts 100+ types of observations from every meeting, among them decisions, action items and risks, each linked to where it came from, and connected tools such as Slack and Teams add the conversation around them. When a decision changes, the new one supersedes the old one, so the record stays usable for years.
- We could run our own MCP server over our meeting notes. Why not?
- Plenty of teams start that way, and for one small team it works. What gets hard comes later: summaries that contradict each other as decisions change, giving different people different answers, and tracing an answer back to what was actually said. In Parallel stores decisions and what they replaced, grants access at four depths, and links every answer to its source. Our build-it-yourself comparison goes through it point by point.
- Does this lock us into In Parallel?
- The opposite. MCP is an open protocol; your agents keep working the same way whatever else changes in your stack. We win by being the best context, not by being hard to leave.
- What does it cost?
- MCP access is included in every plan — from €29 per user per month. Building the context starts with the notetaker: it is the entry point.
Your AI isn’t the bottleneck. Feed it.
The tools are already good. Give them the context they’re missing and every one of them gets better on the same day.
- One endpoint per workspace
- Four access levels, re-checked every call
- Open protocol, no lock-in