GTM-engineering deep dive · MCP + Close · 2026

Close + Claude via native MCP — Dynamic Client Registration, Close-Scope gating, and Chloe AI in one conversation

Close ships a hosted MCP server at https://mcp.close.com/mcp with HTTP Streamable transport, OAuth 2.0 + Dynamic Client Registration (no manual OAuth-app provisioning), and the Close-Scope header for sharp write-surface gating. For GTM engineers and inside-sales operators driving daily orchestration from an AI client, this is the cleanest LLM-native CRM shape for phone-led motions in 2026 — and the Chloe AI integration is the second-order advantage no other MCP-shipping CRM matches.

This page is the setup walkthrough plus five concrete workflows we run in our own RevOps motion, the Close-Scope header explained in operator terms, and the multi-MCP orchestration pattern (Close + Apollo + Smartlead in one Claude session) that collapses the cross-tool friction Zapier was built to bridge.

MCP endpoint
mcp.close.com/mcp
HTTP Streamable + OAuth 2.0 + DCR
Setup time
~60 sec
Dynamic Client Registration handles OAuth provisioning
Entry tier
$9/user/mo
Solo plan, MCP included
Write gating
3-level
read / write_safe / write_destructive

TL;DR

Want to try Close?

Wire Close into Claude in 60 seconds and ship the LLM-native inside-sales motion

Solo $9/user/mo includes the bundled Power Dialer + SMS + Chloe AI + native MCP. Dynamic Client Registration means the OAuth flow is effectively one-click. Cleanest LLM-native CRM shape for phone-led motions in 2026.

Start with Close →Affiliate link — StackSwap earns a commission if you sign up for Close. We only partner with tools we'd recommend anyway.

What makes Close MCP structurally different

Most MCP-shipping CRMs handle the write surface with one of two shapes: full trust (Apollo, HubSpot — you authenticated, you have full access) or per-call confirmation (Attio — every write fires a prompt in the client). Close split the surface a third way: the Close-Scope header gates the entire session into read / write_safe / write_destructive, and the LLM operates inside that scope for the duration.

The operator benefit is bounded blast radius. An over-eager agent loop running under write_safe can create 50 unnecessary leads, but it can't delete the existing 50. The damage is bounded by the scope you connected under — not by the LLM's enthusiasm or your client's per-call confirmation UX. This is the sharpest write-permission model on any MCP-shipping CRM in 2026.

Plus Dynamic Client Registration on the OAuth side, which means MCP clients (Claude, Cursor, n8n, VSCode) register themselves automatically with the Close OAuth server — no "go to Close admin, create an OAuth app, paste the client ID into the MCP config" step. Effectively one-click connect once you're set up.

Five concrete Claude + Close workflows you can ship today

1. End-of-day pipeline summary with Chloe context

Daily prompt: "Summarize today's call activity from Chloe — which deals moved forward, which stalled, which accounts had negative signals." Claude reads via MCP all of today's Chloe summaries plus the underlying lead + opportunity context, returns a structured summary in chat, and (with a follow-up) drafts re-engagement copy for the stalled deals grounded in the actual call content. Under write_safe scope, Claude can queue the re-engagement messages for human review the next morning.

2. Multi-entity follow-up drafting

Drop a list of 20 accounts that had calls this week. Claude reads person + lead + opportunity + Chloe call summary for each in a single agent turn, drafts personalized follow-up emails that reference both the relationship history and the specific call content. The Close-Scope write_safe lets the agent log the drafts as activities pending human send.

3. Multi-MCP outbound orchestration with Apollo + Smartlead

Configure Claude with Apollo MCP, Close MCP (under write_safe), and Smartlead MCP. Workflow: "Source 25 VP Sales prospects at B2B SaaS companies 50-200 employees from Apollo, create them as leads in Close with the appropriate pipeline + custom fields, and push them into the Smartlead Q2 outbound campaign with personalization grounded in company context." End-to-end, one conversation, no Zapier.

4. On-demand Smart View construction

Describe a Smart View in natural language ("leads where the AE hasn't logged a touch in 14 days, with deal size greater than $25K, in stage Demo Scheduled"). Claude translates into the Close Smart View filter syntax and creates the view via MCP. Eliminates the Smart View builder session that most operators avoid because they don't remember the filter syntax.

5. Chloe-context-driven re-engagement (scheduled task)

Schedule Claude to run daily: pull all of today's Chloe call summaries, identify deals with negative signals (objections, "send me information later", longer-than-baseline silence), draft re-engagement copy grounded in the actual call content, and either queue under write_safe for human review or auto-send with explicit confirmation depending on your risk tolerance.

Setup — Dynamic Client Registration makes this fast

  1. Add the Close MCP server to your client. For Claude Desktop or Claude Code, add an MCP entry pointing at https://mcp.close.com/mcp. Specify the Close-Scope header — start with read for the first session. For Cursor, n8n, VSCode: same endpoint, same header.
  2. Authorize via OAuth. The MCP client opens a browser tab for the OAuth flow — log in to Close, authorize, done. Dynamic Client Registration handles the client-ID provisioning automatically.
  3. Connect with a scoped Close user, not your admin account. Same standard advice that applies to every MCP server with a write surface.
  4. Verify connectivity. "List my recent Close leads." If Claude returns a structured response, you're wired in.
  5. Escalate Close-Scope gradually. read for the first week. Move to write_safe for production agent workflows. Reserve write_destructive for explicit cleanup with active human supervision.

When Close MCP doesn't unlock value

Same honesty as the other MCP integration guides. If your daily orchestration doesn't run through Claude / Cursor / ChatGPT and your team works the Close UI directly (Power Dialer, Smart Views, sequence builder), the MCP layer doesn't add value to your motion. You're paying for capability you won't use — though Close's MCP is included on every tier so the "don't pay for it" framing is moot.

For non-AI-forward operators, evaluate Close on the structural wins that hold regardless of MCP: bundled Power Dialer + Predictive Dialer + SMS + Chloe AI + email under one per-seat contract ($9-$139/user/mo), the inside-sales-execution focus that no broader CRM matches, and the lack of seat-padding that hits HubSpot and Salesforce as you scale past the entry tier.

Want to try Close?

If you drive daily orchestration through Claude, Close MCP is the cleanest LLM-native CRM shape for phone-led motions

Close-Scope gating, Dynamic Client Registration, Chloe AI integration, and $9-$139/user/mo per-seat pricing. The cleanest inside-sales motion in 2026.

Start with Close →Affiliate link — StackSwap earns a commission if you sign up for Close. We only partner with tools we'd recommend anyway.

FAQ

Close MCP supports Dynamic Client Registration, which means the connection is nearly one-click. In Claude Desktop or Claude Code, add an MCP server entry pointing at https://mcp.close.com/mcp; specify the Close-Scope header (start with 'read' for the first session, escalate to 'write_safe' once you have calibration). The MCP client opens a browser tab for the OAuth flow — log in to Close, authorize, done. No manual OAuth-app provisioning required. Same flow for Cursor, n8n, and VSCode. Total time: 30-60 seconds.

No. Close MCP is included on every tier — Solo $9/user/mo, Essentials $35, Growth $99, Scale $139. The MCP surface scales with the Close API entitlement (Essentials and up unlock the full API), and Chloe AI access on the MCP surface scales with the Chloe inclusions of your tier. For solo founders running call-first outbound, Solo at $9 is the cheapest serious MCP-CRM entry point in the category — nothing else with this caliber of MCP write surface gets close to that price.

The Close-Scope HTTP header gates the MCP session's write surface into one of three levels: 'read' (search and retrieve only), 'write_safe' (creates, updates, sends — no destructive operations), and 'write_destructive' (full delete + bulk modification). Set it in your MCP client config when you add the Close server. Start at 'read' for the first week of use to observe how the LLM uses the surface. Escalate to 'write_safe' for production agent workflows once you have calibration. Reserve 'write_destructive' for explicit cleanup sessions with a human actively driving — treat the same as a workspace-admin token.

Five validated workflows: (1) end-of-day pipeline summary — Claude reads today's calls (Chloe summaries), surfaces stalled deals, drafts re-engagement copy under write_safe; (2) multi-entity follow-up drafting — LLM reads person + lead + opportunity + Chloe call summary in one turn, drafts contextual follow-up emails; (3) cross-tool sourcing — Apollo MCP sources contacts, Close MCP creates leads under write_safe, sequencing MCP runs cadences, all in one Claude conversation; (4) Chloe-context-driven re-engagement — scheduled daily Claude task pulls Chloe summaries, identifies negative signals, drafts re-engagement, queues for human review; (5) on-demand Smart View construction — describe a Smart View in natural language, Claude builds it via the Close API. Workflows we'd avoid: anything under write_destructive scope without active human supervision.

Yes — this is the multi-MCP outbound pattern. Configure Claude with Close MCP, Apollo MCP, and Smartlead MCP (community-built by LeadMagic, Smartlead's official partner). Claude sources contacts from Apollo, creates leads in Close under write_safe, runs email cadences via Smartlead, watches reply detection across both Smartlead and Close (calls + SMS), and routes accepted prospects to the appropriate Close opportunity stage. The composable shape eliminates the Zapier middleware bill and the per-tool tab-flipping.

Order-of-magnitude friction difference. With Zapier: AI client → webhook → Zapier → Close REST API → response back through Zapier. Each hop adds latency, Zapier seats are a monthly line item, and Zapier maintains the Close integration so schema changes propagate on Zapier's timeline. With Close MCP: AI client → Close MCP → response. One hop. Sub-second. Close maintains its own MCP server, so schema updates ship coordinated with the MCP surface. No middleware bill. Same logic applies for n8n, Make, Workato.

Limited. If your daily orchestration doesn't run through Claude / Cursor / ChatGPT and your team works the Close UI directly (Power Dialer, Smart Views, sequence builder), the MCP layer doesn't add value to your motion. You're paying for capability you won't use — though Close's MCP is included on every tier so the 'don't pay for it' framing is moot. For non-AI-forward operators, evaluate Close on the structural wins that hold regardless of MCP: the bundled Power Dialer + Predictive Dialer + SMS + Chloe AI under one per-seat contract, which alone justifies the $99/user/mo Growth tier vs stacking HubSpot + Aircall + Salesloft + an AI notetaker at 3-4x the cost.

Officially supported by Close. The endpoint is at https://mcp.close.com/mcp (Close's domain), Close runs the server, Close handles the OAuth + Dynamic Client Registration flow, and the docs live at help.close.com/docs/mcp-server. Schema changes, OAuth scope definitions, and the Close-Scope header model are all first-party. There's a real support channel when something breaks. Compare against community-MCP wrappers — those work but the maintenance burden is yours.

Related reading

Canonical URL: https://stackswap.ai/close-mcp-claude-integration. Disclosure: StackSwap is a Close affiliate.