CallGrid MCP: Connect to Claude (Desktop)

Overview

The CallGrid MCP server lets Claude read your CallGrid account directly, so you can ask questions like "which campaigns are live right now?" or "why are my bids getting rejected?" and get answers without logging into the dashboard.

MCP (Model Context Protocol) is the open standard that connects AI assistants to outside tools. CallGrid runs a hosted MCP server, so there is nothing to install and no code to write. You add one URL and you are done.

This guide covers Claude on the desktop app. For the ChatGPT version, see CallGrid MCP: Connect to ChatGPT.

Connection details

Name: CallGrid

Description: CallGrid MCP Server

Server URL: https://mcp.callgrid.com/mcp

OAuth setup: None required

Setup

Setup takes about 30 seconds.

Step 1: Open Connectors

In the Claude desktop app, open your settings and go to Connectors.

Step 2: Add the custom connector

1. Click Add custom connector.
2. Fill in the fields:
3. • Name: CallGrid
   • Description (optional): CallGrid MCP Server
   • Remote MCP server URL: https://mcp.callgrid.com/mcp
3. Leave the OAuth fields blank. CallGrid does not need any OAuth client ID or secret.
4. Click Add.

Claude pulls in the CallGrid icon automatically, so the connector looks correct in your list right away.

Step 3: Authorize your account

The first time you ask CallGrid something, Claude prompts you to connect and authorize access to your CallGrid organization. Approve it once and you stay connected for future chats.

If you manage more than one CallGrid organization, tell Claude which one to use (for example, "use my Acme org"). Claude will use that organization for the rest of the conversation.

Step 4: Start asking

Make sure the CallGrid connector is enabled for your conversation, then ask away. See the examples below.

What you can do with it

The MCP gives Claude live read access to your account. Here are prompts that map to what it can pull today.

Account and structure

• "Which CallGrid organization am I connected to?"
• "List my active campaigns."
• "What traffic sources feed my insurance campaign?"
• "Show me my buyers and their current status."
• "Who are my publishers and vendors?"
• "What routing destinations are set up?"

Numbers and tags

• "How many tracking numbers do I have, and which campaigns use them?"
• "What tags are configured on my account?"

Performance and diagnostics

• "Give me a performance summary for last week."
• "Why are my bids getting rejected? Show the bid stats."
• "How are my pings performing?"
• "What changed on my account recently?" (version history)
• "List the webhooks on my campaigns so I can check one is firing."

Because Claude understands context, you can stack follow ups: ask for a summary, then "now break that down by buyer," then "which of those had the lowest connect rate." No report builder, just questions.

Advanced prompts: reports, scorecards, and optimization plans

Claude does not just fetch numbers. It can pull your data and then reason over it, so you can ask for finished work product: HTML reports, charts, scorecards, and full optimization plans with revenue estimates. Paste any of these, swap in your own specifics, and go.

Build an optimization report with charts

Make an HTML report with graphs on my top performing campaign, plus a list of improvements I could make. Propose specific setting changes, and include charts that estimate what each change could improve if I made it, with the goal of more revenue.

Diagnose bid rejections and quantify the loss

Pull my bid stats for the last 7 days, group the top rejection reasons, and give me a prioritized fix list with an estimate of the calls and revenue I am losing to each.

Score my buyers and flag the laggards

Rank my buyers by conversion rate and revenue per call. Flag the underperformers, and draft talking points I can use to renegotiate the bottom three.

Compare traffic sources and tell me where to spend

Compare my traffic sources by billable rate and revenue per call in a chart, then tell me which vendors to scale and which to cut.

Map the ping to call funnel

Show me the funnel from pings to bids to accepted calls, point to the biggest drop off, and chart it.

Run a webhook health check

Check the webhooks on all my campaigns. List any that look misconfigured or are missing a postback for converted events, and tell me exactly what to add.

A note on estimates: when the assistant projects revenue lift, those are estimates built from your own data, not guarantees. Treat them as a prioritized starting point, then validate with a small test before you roll a change out.

What it can and cannot do

Today the CallGrid MCP is read only. It surfaces and analyzes your data. It does not place calls, create or edit campaigns, move numbers, or change any settings. Asking it to "pause campaign X" will not change anything in your account.

That makes it safe to hand to anyone on your team who needs answers but should not be editing the account. Write actions may come later. For now, think of it as instant reporting and a second set of eyes on account health.

Troubleshooting

• The connector was added but Claude says it has no CallGrid tools. Open the conversation's connector controls and confirm CallGrid is turned on for that chat.
• Claude asks which organization to use. That is expected when your login has access to more than one org. Name the one you want.
• You get an authorization prompt every time. Approve the connection once and let it complete. If it keeps prompting, remove the connector and add it again with the exact URL above.
• Numbers look stale. Ask Claude to pull fresh data ("re-check the latest"), since it answers from the most recent pull in the conversation.