# For Engineering · ClosedLoop AI > Product intelligence in your editor via MCP. Your coding agents already write the code. Give them what your customers actually said, connect once. --- [Skip to main content](#main-content)For engineering # Customer truth, in the editor . **The right thing, at the right scope, before the ticket is written.**Your coding agent asks; the evidence answers. Your AI agents already write the code. They are missing one thing: what your customers actually said. Connect the ClosedLoop AI MCP and Claude Code, Cursor, Windsurf, or Zed can query every customer conversation in plain language. Answers come back with quoted evidence and a source link. No dashboard. No browser tab. No context switch. [Start free](https://app.closedloop.sh/auth?mode=signup)[Book a demo](https://closedloop.sh/contact) No credit card. Authorize in your browser on the first call. In the editor MCP > how many accounts hit the export bug, and what exactly did they say? Quotes Accounts Source links Answers fill from your own calls, tickets, and CRM once a source is connected. Cited, never asserted. What changes for engineering ## Ship the right thing, scoped from evidence. E-01 Scope with evidence "Is this real? How many accounts?" answered in seconds, with the quotes attached. E-02 Skip the CS round-trip No pinging CS and waiting until Tuesday. The conversation history is queryable directly. E-03 Stay in flow The answer lands next to the code, inside the agent you already use. E-04 Close the loop when you ship When the fix ships, the customers who asked surface automatically for CS to notify. The problem ## You don't have a coding problem. You have a requirements problem. Four patterns we hear from every engineering team we talk to. If any of them sound familiar, the answers start the day you connect a source. P-01 You build features blind PM hands you a ticket. You ship it. Six months later the support queue says it doesn't solve what the customer actually meant. P-02 Scoping eats half the work "Is this a real bug?", "how many accounts hit this?", "does anyone use this old endpoint?" The questions before the work are the work. P-03 CS is a bottleneck You ping CS to ask "is this real?" CS pings the account. The account responds Tuesday. Your ticket sits. P-04 Dashboards are the wrong shape You don't want a dashboard. You want the answer in the editor, next to the code, while the question is still open. How it works ## How it slots into your existing flow. No new tool to learn. The MCP registers as a tool in your existing agent. Your existing prompts get smarter. - 01 #### Add the MCP For a US workspace, run `claude mcp add --transport http closedloop-ai https://mcp.closedloop.sh`. Use `https://eu.mcp.closedloop.sh`for an EU workspace, then type `/mcp`and authorize in your browser. - 02 #### Query in plain language `How many customers reported the SSO bug?`Answers come back in seconds with the customers, the quotes, and the revenue at stake. - 03 #### Scope with evidence When you write the ticket, the agent already cites real customer quotes. The PR description writes itself. - 04 #### Ship-notify when it lands Tell ClosedLoop AI what shipped and it surfaces every customer who asked, with drafts ready for CS. You don't have to remember who to tell. Where it runs ## Built for editors that speak MCP. Same MCP. Different surface. Pick the one your team already lives in. [Editor Claude Code Drop the MCP into your Claude config. Conversations grounded in your customer data, with full agentic coding intact. Read more](https://closedloop.sh/integrations/claude)[Editor Cursor Add via Settings, then MCP. Customer evidence inline in the buffer where you write the PRD and the ticket. Read more](https://closedloop.sh/integrations/cursor)[Loop Linear Closed-loop context: your tickets carry the customer evidence behind them, and when the work ships, every customer who asked surfaces for CS. Read more](https://closedloop.sh/integrations/linear)Proof ## Proof gets measured, never manufactured. Two revisions are open for this page. Each lands here only when it clears the standard below: real, attributable, approved. An empty slot beats a manufactured one, so these stay open until then. Rev A · Corpus metric In verification An attributed number from our own analyzed corpus, for example time from question to cited answer. Real and recomputable, or nothing. Accepts when Attributed to our corpus and independently recomputable. Rev B · Engineering quote Awaiting approval A verbatim quote from an engineering leader running the MCP, with a real name, title, and company. Approved, or nothing. Accepts when Named person, title, company, and written consent on file. FAQ ## The questions engineering teams actually ask. If yours isn't here, support@closedloop.sh is the fastest path. We answer within a business day. Which MCP clients are supported? Claude Code (desktop and CLI), Cursor, Windsurf, Zed, Continue, VS Code with Roo or Cline. Any MCP-compliant client works, the MCP speaks the standard protocol. How is authentication handled? Per-user OAuth on the first call. Tokens are scoped to your workspace. Audit log on every MCP call. SSO and SAML on Enterprise. Do I have to install a server? No. The MCP is a hosted endpoint: `https://mcp.closedloop.sh`(or `eu.mcp.closedloop.sh`). Your client connects over HTTP with OAuth; there is nothing to run or patch on your side. What about data residency? EU or US, locked in at workspace creation. We don't train models on your data. SOC 2 Type II attested, ISO 27001:2022 certified. Enterprise adds full audit access. Can the agent write to my tools? MCP tools read your workspace and answer with cited evidence; they don't write to your PM tool or repo. Ticket drafts with evidence attached happen in ClosedLoop AI, where you review before anything moves. ## See it work on your data this week. [Start free](https://app.closedloop.sh/auth?mode=signup)[Book a demo](https://closedloop.sh/contact) Free forever · No credit card · Connect a source and get this Monday's brief. --- ## More Information - Website: https://closedloop.sh - Documentation: /docs - Pricing: https://closedloop.sh/pricing - Contact: https://closedloop.sh/contact