- The Deep View
- Posts
- Why MCP is the secret weapon for top data teams
Why MCP is the secret weapon for top data teams

Good morning.
Welcome to this special edition of The Deep View, brought to you in partnership with Cognition.
It's 4 PM on a Thursday, right before the end of the quarter. Your PM slacks the team asking why user engagement dropped 15% yesterday. But everyone’s heads down sprinting. Your engineer suggests checking a few different dashboards. Someone mentions a recent deployment. No one has time to root through databases to find the answer. The questions pile up.
Here's the thing about data in most startups and companies: it's scattered everywhere. Product engineers instrument events. Data engineers build complex ETL pipelines. Analysts create models. Finance crunches revenue numbers.
Even basic questions like “why is revenue up for this cohort” can require connecting many dots across systems and codebases.
Where agents come in
Cognition (makers of Devin and Windsurf) and Eight Sleep have taken care of their analytics backlogs by teaching agents to be their dedicated always-online data analysts. Not SQL-writing GPT wrappers, but true agents that understand their entire data ecosystem—from raw instrumentation code to dashboards.
Full context understanding: Their agents (Devin) search through product code, ETL pipelines, dbt models, and documentation to understand not just what the data says, but where it comes from and how it's computed.
Multi-step analysis: The most important data questions rarely have simple answers. They require multiple follow-up queries, validation, cohort comparisons, and data visualization. Agents excel at these long-running workflows that can’t be done in a single SQL query.
Tribal knowledge: Devin picks up knowledge about your company’s analytics workflows, helping it get better over time.
With Devin's help, Eight Sleep reports they're shipping 3x as many data features and investigations per week, with their queue of ad-hoc data requests near zero. Cognition estimates their team saves 150+ hours per month on routine data tasks.
MCP: the protocol that makes it easy
For best results, connect your agent to your databases via Model Context Protocol (MCP)—an open standard that's quickly becoming the "universal adapter" for agents.
Before MCP became popular, connecting an AI to your data stack meant building custom integrations for every tool. PostgreSQL? Custom code. Datadog? Another integration. Sentry? Integration hell.
MCP changes this by creating a standardized way for AI agents to connect to any tool or service. It's like having one charging cable that works with every device.
With Devin’s MCP Marketplace, you can easily give Devin carefully-controlled access to MCP servers for Datadog, Snowflake, Looker, Redshift, Metabase, Sentry, Notion, and any other data source with an MCP server.
Tips from Cognition and Eight Sleep
Clean data architecture and context: If your data is a mess, no agent can save you. Companies with well-structured data models see much better results. Eight Sleep adds helpful notes to their dbt yaml files using the meta field. This helps agents figure out what they’re looking at.
Guardrails: Just like humans, agents need clear permissions. What data can they access? What operations are allowed? Cognition recommends giving agents read-only access.
Trust but verify: Cognition recommends asking agents for their thought process and sequence of queries so humans can verify that everything checks out. Internally, they often ask Devin for Metabase playground links so they can go in and inspect the SQL. Eight Sleep notes that providing feedback to the agent when it’s wrong helps it improve and learn from its mistakes.
Getting Started
If you’re interested in a battle-tested solution, Devin is available starting at $20. Once you’re in, visit the MCP Marketplace in Settings and check out Cognition’s tutorial on setting up Devin for analytics.
Beyond analytics, you can also give Devin access to MCP servers for tools like Linear, Figma, and more.
Windsurf, the privacy-first enterprise-ready agentic IDE, also supports MCP. Used by teams at Chase, VMware, and Intel, Windsurf has all the latest AI tooling as well as support for ZDR, SOC II, HIPAA, FedRAMP, and more.

Take The Deep View with you on the go! We’ve got exclusive, in-depth interviews for you on The Deep View: Conversations podcast every Tuesday morning.

If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.