# Claude Code MCP · ClosedLoop AI

> Claude codes better when it understands the real user problem, not just the ticket. Connect over HTTP and authorize with OAuth. Product insights inside Claude Code.

---

[Skip to main content](#main-content)

## MCP for Claude Code


- [Overview](#top)
- [What it looks like](#demo)
- [What you can ask](#ask)
- [Connect](#install)


## More


- [MCP docs](https://closedloop.sh/docs/mcp-server/overview)
- [API reference](https://closedloop.sh/developers/api)
- [Browse integrations](https://closedloop.sh/integrations)

Product context in Claude Code


# Customer evidence, inside Claude Code.


Claude codes better when it understands the real user problem, not just the ticket. Connect once with OAuth, then ask Claude what to build, which bugs matter, and what customers are saying.
$ `claude mcp add --transport http closedloop-ai https://mcp.closedloop.sh`Copy [Read docs](https://closedloop.sh/docs/mcp-server/overview)

- Claude Code
- OAuth 2.1
- Free


## What it looks like


Ask Claude to scope a feature with ClosedLoop AI before implementation. It queries insight counts, deal blockers, and verbatim quotes without sending you to another tool.
Claude Code ClosedLoop AI connected
> use ClosedLoop AI to scope the data export feature

ClosedLoop AI / **Customer evidence: data export**
**Role-specific export templates**. 6 insights "Manual reporting takes ~3h/week, hundreds of rows nobody reads" **Background export for large datasets**. 4 insights Deal blocker "We export 50k rows and it just times out" **CSV export with filters**. 8 insights "I just need to export what I'm looking at, not the whole table"
**18 insights from 12 customers.**The main pain is large datasets timing out and no way to filter before exporting.
Background async export with download link CSV with current filters applied (8 asked for this) Progress indicator: "I had no idea if it was running" Skip Excel/PDF for now, only 2 mentions
Want me to start with the background job implementation?


## What you can ask Claude


Hundreds of decisions per feature: API design, edge cases, what to ship now versus later. Claude makes better calls when it knows what users actually said.
Implement

### Build with context


Ask Claude to query real user pain, workarounds, and expected behavior before it writes code.
"use ClosedLoop AI before implementing export" Plan

### Generate a PRD


Get a full product requirements doc backed by real customer quotes, insight counts, and workarounds. Not opinions, evidence.
"write a PRD for data export using customer evidence" Review

### Review a tech spec


Ask Claude to compare a proposed design with customer evidence before you commit to the implementation.
"review this export spec against customer evidence" Scope

### Scope v1 from evidence


Claude sees which features have 12 insights and which have 1. Your v1 includes what users asked for and skips what they didn't.
"what's the minimum scope for export?" On demand

### Evidence while you work


Ask for customer evidence in the same conversation where you plan, scope, and build the feature.
"ground this plan in customer evidence" Search

### Search your feedback


Ask Claude about any topic and it searches your product insights: pain, workarounds, severity, deal blockers. Not a Jira summary.
"what are customers saying about search?"
Setup 2 minutes


## Connect in minutes


Add the hosted endpoint, authorize your workspace, and start asking questions.
Used a legacy installer?
Remove its user-scoped client before connecting: run `claude mcp remove --scope user closedloop-ai`for the former macOS/Linux installer, or `claude mcp remove --scope user closedloop`for the former Windows installer.


- ### Add the MCP


Run the command for your workspace region. By default, Claude Code adds the hosted connection to the current project.
$ `claude mcp add --transport http closedloop-ai https://mcp.closedloop.sh`Copy
For an EU workspace, use `https://eu.mcp.closedloop.sh`.
- ### Authorize your workspace


Type `/mcp`in Claude Code, open the authorization link, then sign in and choose the workspace you want to connect.
- ### Manual config (optional)


For clients that use a JSON configuration file, add only the hosted endpoint:
json MCP client config
```
`{ "mcpServers" : { "closedloop-ai" : {
      "url": "https://mcp.closedloop.sh"
    }
  }
}`
```
- ### Ask Claude


Start exploring your product insights and opportunities. Claude now has read access to the workspace you authorized.
OAuth in your browser
Your client discovers the authorization flow automatically. Sign in, choose a workspace, and approve the connection; your client manages the resulting access token.


## Start building from customer evidence

[Setup Connect with OAuth → Choose your client, add the endpoint, and authorize your workspace.](https://closedloop.sh/docs/mcp-server/overview)[Reference Explore the tools → See every available tool, parameter, and return value.](https://closedloop.sh/docs/mcp-server/tools)

---

## More Information

- Website: https://closedloop.sh
- Documentation: /docs
- Pricing: https://closedloop.sh/pricing
- Contact: https://closedloop.sh/contact