PipeCrush APIMCP ServerLive

Connect your AI agents to PipeCrush.

PipeCrush runs a remote MCP server at api/mcp, and the same tools ship as a local server you can run with npx. Point Claude Code, Claude Desktop, Cursor, VS Code, or Gemini CLI at either one with your API key, and your agent can look up records and make updates in your workspace.

Get your API key Explore the REST API

76 tools, no code to write

Optional read-only keys

Claude

Connected to PipeCrush

Which leads came in this week, and do any of them need a follow-up?

list_leadscreated_after: "monday"

create_tasktitle: "Follow up with Jane"

Twelve leads came in this week. Two have no activity since Tuesday — I've created a follow-up task for Jane at Acme and can draft one for the other if you'd like.

How it works

Three steps, about ten minutes, no code.

Create an API key

In PipeCrush, go to Settings, then API Keys, and click Create key (admin role required). Choose a read-only key if the agent only needs to look things up. Copy the key immediately — it is shown only once.

Choose remote or local

Point your MCP client at https://www.pipecrush.tech/api/mcp with your API key in the Authorization header, or run the server locally with npx -y @pipecrush/mcp and your key in an environment variable. Claude Code, Cursor, VS Code, and Gemini CLI support both; Claude Desktop uses the local path.

Start asking

Restart or reconnect your client and ask about your pipeline. Claude discovers the 76 PipeCrush tools automatically and uses them to look up and update your records.

No API key needed

Connect with OAuth

claude.ai, Claude Desktop, ChatGPT, and Claude Code can all connect to PipeCrush by signing in — no key to create, copy, or store.

claude.ai

Settings → Connectors → Add custom connector, then enter https://www.pipecrush.tech/api/mcp. claude.ai opens PipeCrush sign-in for you.

Claude Desktop

Settings → Connectors → Add custom connector, then enter the same URL. Desktop can also run the server locally with npx and an API key — see below.

ChatGPT

Turn on developer mode, then add a custom connector pointing at https://www.pipecrush.tech/api/mcp.

Claude Code

claude mcp add --transport http pipecrush https://www.pipecrush.tech/api/mcp

The flow is the same everywhere

Sign in to PipeCrush, choose Read-only or Read + write access, and you're connected. Revoke access any time under Settings → Connected AI apps — the app's very next request fails immediately.

API keys aren't going anywhere — they're still the right choice for scripts, servers, and the local npx -y @pipecrush/mcp path below, where there's no human sign-in step.

Connect your MCP client

Two ways to connect: point a client at https://www.pipecrush.tech/api/mcp with your API key as a bearer token in the Authorization header, or run the server locally with npx -y @pipecrush/mcp and your key in a PIPECRUSH_API_KEY environment variable. Most clients below support both.

Claude Code

Works today

Anthropic's terminal coding agent. The fastest way to try this.

Remote (recommended) — run this

claude mcp add --transport http pipecrush https://www.pipecrush.tech/api/mcp \
  --header "Authorization: Bearer pc_live_your_key_here"

Local (npx) — run this

claude mcp add pipecrush --env PIPECRUSH_API_KEY=pc_live_your_key_here \
  -- npx -y @pipecrush/mcp

Cursor

Works today

The AI code editor. Add the server to your mcp.json.

Remote (recommended) — add this to your MCP config

{
  "mcpServers": {
    "pipecrush": {
      "url": "https://www.pipecrush.tech/api/mcp",
      "headers": {
        "Authorization": "Bearer pc_live_your_key_here"
      }
    }
  }
}

Local (npx) — add this to your MCP config

{
  "mcpServers": {
    "pipecrush": {
      "command": "npx",
      "args": ["-y", "@pipecrush/mcp"],
      "env": {
        "PIPECRUSH_API_KEY": "pc_live_your_key_here"
      }
    }
  }
}

VS Code

Works today

With GitHub Copilot agent mode. Add it to .vscode/mcp.json.

Remote (recommended) — add this to your MCP config

{
  "mcpServers": {
    "pipecrush": {
      "url": "https://www.pipecrush.tech/api/mcp",
      "headers": {
        "Authorization": "Bearer pc_live_your_key_here"
      }
    }
  }
}

Local (npx) — add this to your MCP config

{
  "mcpServers": {
    "pipecrush": {
      "command": "npx",
      "args": ["-y", "@pipecrush/mcp"],
      "env": {
        "PIPECRUSH_API_KEY": "pc_live_your_key_here"
      }
    }
  }
}

Gemini CLI

Works today

Google's terminal agent. Add it to your settings.json.

Remote (recommended) — add this to your MCP config

{
  "mcpServers": {
    "pipecrush": {
      "httpUrl": "https://www.pipecrush.tech/api/mcp",
      "headers": {
        "Authorization": "Bearer pc_live_your_key_here"
      }
    }
  }
}

Local (npx) — add this to your MCP config

{
  "mcpServers": {
    "pipecrush": {
      "command": "npx",
      "args": ["-y", "@pipecrush/mcp"],
      "env": {
        "PIPECRUSH_API_KEY": "pc_live_your_key_here"
      }
    }
  }
}

Claude Desktop

Works today

The Claude desktop app. Connect with OAuth, or run the server locally with npx.

OAuth (recommended) — follow these steps

Settings → Connectors → Add custom connector → paste https://www.pipecrush.tech/api/mcp → sign in to PipeCrush → choose Read only or Read + write → Allow.

Local (npx) — add this to your MCP config

{
  "mcpServers": {
    "pipecrush": {
      "command": "npx",
      "args": ["-y", "@pipecrush/mcp"],
      "env": {
        "PIPECRUSH_API_KEY": "pc_live_your_key_here"
      }
    }
  }
}

ChatGPT

Works today

Custom connectors in developer mode. Connect with OAuth — no API key or local process needed.

OAuth (recommended) — follow these steps

Settings → Connectors (developer mode) → Add custom connector → paste https://www.pipecrush.tech/api/mcp → sign in to PipeCrush → choose Read only or Read + write → Allow.

claude.ai

Works today

Custom connectors on the web. Connect with OAuth — no API key or local process needed.

OAuth (recommended) — follow these steps

Settings → Connectors → Add custom connector → paste https://www.pipecrush.tech/api/mcp → sign in to PipeCrush → choose Read only or Read + write → Allow.

Not yet connectable

Blocked by the client, not by PipeCrush — see each card for why.

Gemini Spark

Not yet connectable

Custom apps in the Gemini web app.

Spark's custom MCP connector is built on OAuth 2.0, and custom apps currently require a personal Google Account rather than a work or school account.

The API key travels in every request — protect it like a password. If a key leaks, revoke it in Settings → API Keys and create a new one; traffic on the old key stops immediately.

76 tools, one per operation

Every tool maps to one operation on the PipeCrush REST API: 24 record tools — list, create, fetch, or update across all six record types — plus 52 action and context tools that send email and broadcasts, enroll a lead in a sequence or campaign, place a call, send a text, book or reschedule an appointment, reply in an inbox thread, read notes, activity, transcripts, and metrics, and manage webhooks. There are no delete tools because the API has no delete endpoints.

The action tools do the thing, not queue it for review — an agent calling send_email sends a real email. Give an agent a read-only key unless you intend that.

Record type	list	create	get	update
contacts	list_contacts	create_contact	get_contact	update_contact
leads	list_leads	create_lead	get_lead	update_lead
customers	list_customers	create_customer	get_customer	update_customer
tasks	list_tasks	create_task	get_task	update_task
tickets	list_tickets	create_ticket	get_ticket	update_ticket
deals	list_deals	create_deal	get_deal	update_deal

The 52 action and context tools

Grouped by the job they do. Every list_* and get_* tool, plus search_records, is available to a read-only key; the rest need write access.

Area	What it does	Tools
Email	Send a one-off email from a connected account and fetch the sent record.	list_email_accountssend_emailget_email
Inbox	Read email conversations from connected inboxes and reply in-thread.	list_inbox_threadsget_inbox_threadreply_to_thread
Sequences	Enroll a lead before launch, stop or resume one lead, and pause or resume a whole sequence.	list_sequencesupdate_sequenceenroll_leadget_sequence_enrollmentupdate_sequence_enrollment
Campaigns	Enroll a lead into an evergreen email campaign, or unsubscribe them from it.	list_campaignsenroll_lead_in_campaignget_campaign_enrollmentupdate_campaign_enrollment
Broadcasts	Create a one-off email-marketing blast, send or schedule it, watch its counters, and cancel it while it is still queued.	list_broadcastsget_broadcastcreate_broadcastsend_broadcastcancel_broadcast
Calls	Place a real outbound call, follow its status, and read the transcript afterward.	list_phone_numberslist_callsplace_callget_callget_call_transcript
Texts	Send an SMS or MMS from a 10DLC-registered number and follow the thread.	list_smssend_smsget_sms
Appointments	Check open slots, book, reschedule, and cancel meetings on a calendar.	get_availabilitylist_appointmentsbook_appointmentget_appointmentreschedule_appointmentcancel_appointment
Notes & comments	Read and add notes on leads and deals; read and add ticket comments, public or internal.	list_lead_notescreate_lead_notelist_deal_notescreate_deal_notelist_ticket_commentscreate_ticket_comment
Lookups & reporting	Find your pipelines and stages, search across every record type, read a lead's full activity timeline, and pull sales funnel metrics.	list_pipelinessearch_recordsget_lead_activityget_sales_metrics
Webhooks	Register HTTPS endpoints that receive signed events, test them with a ping, rotate secrets, and inspect every delivery attempt.	list_webhook_event_typeslist_webhooksget_webhookcreate_webhookupdate_webhookrotate_webhook_secrettest_webhooklist_webhook_deliveries

Built so you can trust an agent with it

An AI agent with access to your CRM deserves the same scrutiny as a new hire. Here is what the server does — and refuses to do — with that access.

Your key stays in one header

The endpoint reads your API key from the Authorization header on each request, sends it only to PipeCrush over TLS, and scrubs it from every log line and error message.

No delete tools

The API has no delete endpoints and the server exposes no delete tools. The worst a confused agent can do is create or edit a record — never destroy one.

Read-only keys

Create a read-only key in Settings → API Keys and only the 42 lookup tools are exposed. Running the server locally instead? Set PIPECRUSH_READ_ONLY=true in its environment for the same effect. Either way, Claude can answer questions about your data without being able to change it.

Isolation enforced server-side

Every request runs through the same tenant isolation as the rest of PipeCrush. A key can only ever see its own organization's data — the agent cannot broaden that.

Bounded by your key's own permissions

Every write the agent makes runs under your API key against your own workspace — so even if it is misled by text inside a record it just read, nothing it does can reach beyond what that key is already allowed to touch. Give agents a read-only key if they only need to look things up.

Frequently asked questions

What is the Model Context Protocol (MCP)?

MCP is an open standard that lets AI assistants like Claude connect to outside tools and data sources. The PipeCrush MCP server speaks that standard, so Claude can read and update your PipeCrush records the same way it would use any other tool.

What do I need to connect Claude to PipeCrush?

Two things: a PipeCrush account with an API key (created by an admin in Settings under API Keys) and an MCP client — Claude Code, Claude Desktop, Cursor, VS Code, or Gemini CLI. Setup is one command or one config block — no code to write.

How do I connect to the PipeCrush MCP server?

Two ways. Point your client at https://www.pipecrush.tech/api/mcp with your API key in the Authorization header — Claude Code has a one-line command, and Cursor, VS Code, and Gemini CLI use a small JSON config block. Or run the server locally with npx -y @pipecrush/mcp and your key in a PIPECRUSH_API_KEY environment variable — this is how Claude Desktop connects, and Claude Code, Cursor, VS Code, and Gemini CLI can use it too. See the setup cards above for the exact commands.

Can Claude delete my records?

No. The PipeCrush API has no delete endpoints, and the MCP server exposes no delete tools. Claude can list, look up, create, and update records — nothing it does through the server can destroy data.

What is a read-only key?

When you create an API key in Settings → API Keys, you can leave write access off. A read-only key only unlocks the 42 lookup tools — Claude can answer questions about your pipeline, read notes, inbox threads, transcripts, and metrics, but cannot create or change anything. Use it for reporting, research, and any agent that only needs lookups.

What data can Claude see?

Exactly what the API key can see: the key acts with the data visibility of the user who created it, and all access stays inside your organization. Record contents are passed to Claude as data for the conversation you're having — nothing is stored by the server itself.

Do MCP tool calls count against my rate limit?

Yes. MCP tool calls share the same limit as the REST API — 120 requests per minute, per API key or per OAuth connection — and each tool call counts as 2 requests, so a connection can make roughly 60 tool calls per minute. Action tools (send email or reply in a thread, place a call, send a text, enroll a lead, book or reschedule an appointment, send a broadcast) also draw against your daily action caps — per key, plus a workspace-wide cap shared across all your keys — on top of the per-minute limit. Broadcast sends are capped at 5 per key per day.

Can I use this with Claude Desktop or the claude.ai web connectors?

Yes. Connect claude.ai and Claude Desktop with OAuth — sign in to PipeCrush and choose Read-only or Read + write access, no key involved. Claude Desktop can still run the server locally with npx -y @pipecrush/mcp and an API key if you prefer that path. ChatGPT's developer-mode custom connectors work the same way, over OAuth.

OAuth vs API key — which should I use?

Use OAuth for AI apps you sign in to interactively — claude.ai, Claude Desktop, ChatGPT, and Claude Code all support it: no key to create, copy, or store, and you can revoke access in one click. Use an API key for anything that runs without a human present — scripts, servers, CI jobs, and the local npx -y @pipecrush/mcp path — where there is no sign-in step to attach the grant to.

Can Claude send a marketing blast or fire off webhooks?

Only with a write-access key or a Read + write OAuth grant. send_broadcast queues a real email-marketing send to up to ten thousand recipients and is capped at 5 per key per day; it also refuses to send without a verified sending domain and a physical mailing address on your organization. Webhook tools register and manage your own endpoints — Claude can create one pointing at your automation platform and send it a test ping, but the signing secret is shown once and never stored by the server.

How do I disconnect an AI app?

Go to Settings → Connected AI apps, find the app, and click Revoke. This takes effect immediately — the very next request from that app fails with a 401 error, the same as deleting an API key.

Put Claude to work in your pipeline

Create a free PipeCrush account, generate an API key, and point Claude Code or another MCP client at the endpoint above.