MCP Integration
MCP (Model Context Protocol) is the way atypica.AI connects to external tools and data sources. Through MCP, the AI can read from and write to systems outside atypica.AI — your internal databases, your proprietary APIs, your custom workflows.
What MCP enables
Without MCP, the AI works within atypica.AI's own capabilities: it uses its built-in tools, the personas library, web search, and the data you upload as attachments.
With MCP, the AI can:
- Pull live data from your internal systems during a research session.
- Trigger actions in your tools based on research findings.
- Read from proprietary knowledge bases the AI wouldn't otherwise have access to.
- Connect to other AI services and route work between them.
When MCP is the right tool
MCP is for enterprise and team-level integration where you need the AI to interact with systems that aren't publicly accessible. Common patterns:
- A consumer brand wants the AI to query its internal sales database while analyzing market trends.
- A financial team wants the AI to read from their proprietary research feed.
- A team wants the AI to trigger downstream actions — push findings into a CRM, file a ticket, update an internal wiki.
If you don't have proprietary systems to connect, MCP isn't relevant — the built-in tools already cover web search, social listening, and document reading.
How MCP is configured
MCP integrations are set up at the team level, not per session. A team admin configures each MCP server the team needs: the URL, authentication headers, and an optional description of what the server does.
Once configured, MCP servers are available to every member of the team during their research sessions. Each session can choose which MCP servers to bring in.
How MCP appears in a session
When MCP is active in a session, the AI's tool palette includes the MCP server's tools alongside the built-in ones. The AI decides when to use them based on the conversation — if you ask "what did our enterprise customers buy last quarter," it'll query the MCP-connected sales database; if you ask "what does Gen Z think," it'll fall back to Scout and web search.
You don't have to remember which MCP server does what. Tell the AI what you want to learn, and it'll reach for the right tool.
Limits
- MCP servers are configured by admins, not by individual users during a session. If you need a new server, ask your admin to set it up.
- MCP requires authentication to your external systems. The AI accesses them with the team's credentials — anyone on the team with session access can trigger those calls.
- MCP integrations add latency and cost. Each external call takes time and counts toward token usage. Don't enable more servers than you actually need.
- MCP doesn't replace the AI's own reasoning. The AI uses MCP tools the way you might use a calculator — for specific data, not for judgment.
Privacy and access
MCP credentials are stored at the team level and used to make requests on behalf of the team. The AI doesn't store or learn from the data it pulls through MCP — it's used for the current session and discarded.
Audit logs show what MCP calls were made and when. Check your team's logs if you need to trace what the AI did.
Example integrations
In practice, MCP is most useful for:
- Internal knowledge — connecting to Notion, Confluence, or proprietary wikis so the AI can reference internal docs.
- Live data — pulling real-time metrics from dashboards, sales systems, or analytics platforms.
- Downstream actions — letting the AI update CRMs, file tickets, or post to internal channels after research.
- Cross-AI orchestration — having atypica.AI call another AI service (or vice versa) as part of a workflow.
If you're unsure whether MCP is the right fit, start by using the built-in tools. Add MCP when you have a specific external system the AI can't reach otherwise.