Apollo.io MCP: how to run Apollo from Claude, ChatGPT, and other AI tools

If you use Apollo and an AI assistant like Claude or ChatGPT, you can now run one from inside the other. You ask in plain language, and the assistant searches Apollo, builds the list, drafts the sequence, and reads the results back, all in your own account. The piece that makes this possible is the Apollo.io MCP.

This guide covers what it is, which AI tools it works with, the three ways to set it up, what costs credits, and which setup fits which job. It’s about Apollo.io, the sales platform. Apollo GraphQL has an MCP server too, and it’s a different product.

What the Apollo.io MCP is

MCP is the standard that lets an AI assistant work inside another tool. Apollo’s version puts your Apollo account in front of your assistant as a set of tools it can call: search for people and companies, enrich a contact, create and fill lists, build sequences and enroll contacts, and pull reporting.

It’s bigger than most people expect. When we counted on 8 November 2026, it listed 96 tools, up from 74 in September. Apollo ships fast, so treat any number here as a snapshot and check what your own connection lists.

Apollo also marks every tool as either read-only or destructive, and the split is worth understanding before you start. Read-only tools are the free ones, like searching for people. Destructive doesn’t only mean “deletes things”. It covers anything that spends credits or changes something live, like enriching a contact or editing a sequence. If your assistant is about to call one of those, that’s the moment to ask what it will cost.

Which AI tools it works with

Apollo lists Claude, ChatGPT, and Perplexity as supported, and says the MCP works on any Apollo plan, the free one included. In each of those tools, you open the connectors or integrations directory, search for Apollo, and sign in with your Apollo account. No API key is needed.

If your tool isn’t on that list but supports MCP, you can still connect it. Paste Apollo’s server address, https://mcp.apollo.io/mcp, into the tool’s MCP settings and approve the sign-in. That covers coding tools like Claude Code and Codex as well. Apollo’s own page keeps the current list.

Three ways to set it up

All three routes reach the same Apollo account with the same data. What changes is how much method comes with the connection.

The connector on its own

This is the route for most people, and it works the same way in every supported tool. In Claude, for example, you go to Customize, then Connectors, find Apollo.io, and sign in. In Claude Code, you add it with one command, claude mcp add --transport http apollo https://mcp.apollo.io/mcp, then run /mcp to sign in. From there you ask for what you need, and the assistant picks the tools.

Apollo’s plugin for Claude Code

If you work in Claude Code, Apollo publishes its own plugin for it. It connects the same MCP and adds five skills: /apollo:prospect, /apollo:enrich-lead, /apollo:sequence-load, /apollo:analytics, and /apollo:gtm-strategist. The last one is the most interesting. It forms a play, builds the audience, and drafts the emails, and it’s disciplined about it: it diagnoses before it searches and it starts with a small test group.

Apollo Operator on top

Apollo Operator is our free, open-source toolkit, and it runs on the same connection in Claude Code and Codex, with a prompt pack for any other AI chat. It adds the method around the tools: sending infrastructure and warmup, list building with independent verification, sequence writing and review, a launch checklist, reporting, and a 15-check account audit. The setup guide takes a few minutes. It sits beside Apollo’s plugin, not in place of it, and you can have both installed.

Apollo has its own agent, and it’s good

Twelve of the tools on the MCP are doors to one thing: Apollo’s own AI agent. You give it an instruction in plain language, it works inside your account, and it comes back with a result. It can find prospects, plan a campaign, draft sequence copy, diagnose a sequence that isn’t working, and answer questions about how Apollo itself works. Apollo has made the same agent available from Slack.

We tested the campaign planner on our own account. It read our company profile in Apollo, planned four campaigns that matched our four real services, sized each audience against live data, and stopped to ask before doing anything. It spent no credits.

So if your question is “who should I target and what should I say”, Apollo’s agent and its strategist skill are a strong place to start, and you don’t need anything from us to use them.

What spends credits, and which credits

Apollo doesn’t have one credit balance. Our account shows eleven separate pools, and spending from one doesn’t touch the others. Two of them tell you most of what you need to know.

• Lead credits are the scarce ones. Our plan has 4,000. Enriching a contact, revealing an email, and searching companies all draw on them.

• AI credits are plentiful. The same plan has 800,000. Apollo prices its own AI writing as close to free and prices the data as the thing you pay for.

The practical rule follows from that. Searching for people is free, so search as much as you like. Enrichment is where the money goes, so do everything cheap first: narrow the search, check what came back, and remove the people you’d never email. Then enrich what’s left. And when anyone tells you how many credits a job will cost, ask which pool.

Where the results go, and why big lists need a different route

Everything the MCP returns passes through your conversation with the assistant. For 20 contacts that’s exactly what you want. For 3,000 it’s a problem, because the results fill the conversation before the work is done.

The numbers make it concrete. One list we pulled, 3,238 people, came to four megabytes of data. That doesn’t fit in a conversation, however you ask. Through Apollo’s command line tool, the same pull went straight to a file and the assistant never had to read it.

The rule we work to is to search wide outside the conversation and act narrow inside it. Pull, check, and trim the big list to a file. Then hand the short final list to the MCP for the steps where you want the assistant asking before it acts, like enrolling contacts into a sequence. Apollo Operator is built around that split, and it’s the main reason a toolkit helps once your lists get large.

What you still have to check yourself

Apollo’s tools are good at finding people and getting them into a sequence. A few jobs sit outside what any connector does for you, and they’re the ones that protect your sending domains.

• Reading what a search returned. A filter states your intent. It doesn’t promise the result. A Founder title filter also matches Founding Engineer, and one search of ours came back as 679 or 91,649 depending on a single setting. The detail is in Apollo filters that inflate your TAM.

• Verifying emails independently before anything sends.

• Removing people you shouldn’t email: customers, open deals, anyone who opted out, and anyone a previous campaign already reached.

• Waiting for warmup. No tool can shorten it.

This is the half of the work Apollo Operator exists for. If you’d rather do it by hand, the Apollo.io setup guide lists every step in order.

Three things that catch people

We’ve run into each of these on our own account, so they’re worth knowing before you start.

1. The connection can expire in the middle of a task. Your assistant suddenly reports that the connector needs authentication. Your account is fine. Reconnect and carry on. For long jobs, Apollo’s docs describe a route that signs in with an API key instead, and that key should be treated like a password.

2. The tool list changes. Between late September and early November it gained 11 tools and dropped several others. If a guide, including this one, names a tool your connection doesn’t list, trust your connection.

3. Editing a live sequence replaces what you don’t send back. An update has to include every step you want to keep. Have your assistant read the sequence first, then change it.

If an agent runs it for you

More teams now hand Apollo to an agent and step back, in Claude Code, Codex, or a tool they built themselves. This section is written for that agent. If you’re a person, you can skip it, or paste it into your agent’s instructions.

• Check access before real work. Call apollo_users_api_profile to confirm the connection and apollo_email_accounts_index to list the mailboxes. If either fails, stop and report it. Neither spends credits.

• Treat the live tool list as the authority. Read the tools your connection lists before you plan. If a tool named in any guide isn’t there, don’t call it and don’t tell the operator Apollo can’t do the job until you’ve checked.

• Stop and ask before any tool marked destructiveHint. It either spends credits or changes something live. Quote the cost first, say the confirmation sentence in the tool’s description as written, and report the mcp_credits block from the result without being asked.

• Name the credit pool. lead_credit is the scarce one. Never report “credits” without saying which pool was spent.

• Keep big pulls out of the conversation. Above about 100 records, pull to a file with Apollo’s command line tool and work from the file. Read the match count first, and narrow the filters before you pull.

• Dedupe by person before you enrich. Pages can overlap, and enrichment charges for every row.

• Read a sequence before you change it. Call apollo_emailer_campaigns_show, then send back every step, touch, and template you want to keep in apollo_sequences_update. Anything left out is deleted.

• Check that a sequence can send. The same show call returns sending_diagnostics, which says whether a schedule is set and whether any email step still needs approval.

• Never activate. Build sequences inactive and enroll contacts. A person turns the sequence on.

• If the connector asks for authentication mid-task, stop. The account isn’t broken. Ask the operator to reconnect, and don’t retry in a loop.

• Never create a test contact with a real email address. Apollo matches it to the existing record and overwrites it.

Which setup to pick

The right setup depends on the job in front of you, and most teams end up using more than one.

If you’d rather have someone run all of it inside your account, that’s Apollo Managed, and we’ve written up how the two routes compare. If you want the account checked first, the audit and 14-day trial is $1,500.