B2B data enrichment for Claude and AI agents.

An MCP server for sales data is a service that exposes a B2B contact database or enrichment engine as tools an AI assistant can call directly. MCP is the Model Context Protocol, described on modelcontextprotocol.io as "an open-source standard for connecting AI applications to external systems," supported by Claude, ChatGPT, VS Code and Cursor among others.

In practice it means an assistant can run a people search, pull a verified email, or push a record to your CRM without you opening another tab. Cleanlist runs one at https://mcp.cleanlist.ai/v1/mcp. So do several competitors: Apollo, ZoomInfo, whose documentation describes a remote server (fetched August 7, 2026).

The sales MCP servers with published first-party documentation in 2026 include Cleanlist, Apollo and ZoomInfo, and the one worth connecting is usually whichever data you already pay for. Apollo has the cheapest entry and states on apollo.io/product/mcp that "Any Apollo plan qualifies, including free". ZoomInfo markets its server as a way to "Connect ZoomInfo data to Claude, ChatGPT, Perplexity, Replit, or custom AI agents". Both fetched August 7, 2026.

Cleanlist runs a remote server at https://mcp.cleanlist.ai/v1/mcp with 30+ tools, free search, pay-for-results enrichment through a 25+ provider waterfall, and signed-quote spend caps. The longer side-by-side, including setup paths and pricing for ten servers, is in the Cleanlist guide to sales and GTM MCP servers.

You connect Cleanlist to Claude by opening Claude, going to Settings, then Connectors, adding a custom connector pointed at https://mcp.cleanlist.ai/v1/mcp, and authorizing it with OAuth. There is no API key to copy or paste and no local process to run: mcp.cleanlist.ai is a remote MCP server, so Claude talks to it over HTTP and the OAuth grant links it to your Cleanlist organization.

It is a custom connector rather than a directory click because Cleanlist is not listed in Anthropic's connector directory yet. Claude reads the authorization server out of the endpoint's OAuth metadata, so the sign-in still happens in a browser window with no key to handle. Once authorized, describe what you need in plain language and Claude picks the Cleanlist tools itself. Revoking from the same Connectors screen cuts the agent off from both your data and your credits immediately.

You connect Cleanlist to Claude Code by running claude mcp add --transport http cleanlist https://mcp.cleanlist.ai/v1/mcp, then running /mcp inside Claude Code and completing the OAuth sign-in. Anthropic documents that syntax for remote HTTP servers and states "Use /mcp to authenticate with remote servers that require OAuth 2.0 authentication" (code.claude.com/docs/en/mcp, fetched August 7, 2026). The Cleanlist endpoint returns 401 until that grant exists, which is how Claude Code knows to prompt you.

Scope is the decision worth making up front. Adding it with --scope local keeps the server on your machine, while --scope project writes it to .mcp.json for everyone on the repo. The OAuth grant is per user but the credit wallet is shared across your whole Cleanlist organization, so a project-scoped server means teammates draw on the same balance.

No. Cleanlist is not in Anthropic's connector directory as of August 7, 2026: claude.com/connectors/cleanlist returns a 404, while comparable B2B data connectors are listed, including Vibe Prospecting at claude.com/connectors/vibeprospecting, plus Apollo and Clay. All four URLs were checked on August 7, 2026: three return 200, ours returns 404.

The practical difference is two clicks. Instead of picking Cleanlist from the directory, you add a custom connector in Claude Settings, then Connectors, and paste https://mcp.cleanlist.ai/v1/mcp. Everything after that is identical: the same OAuth flow, the same tools, the same credits. Anthropic notes that directory connectors run on the same MCP infrastructure as any remote server you add yourself.

Cleanlist documents and tests Claude and Claude Code first, and because https://mcp.cleanlist.ai/v1/mcp is a standard remote MCP endpoint over HTTP with OAuth, any spec-compliant client that supports remote servers can point at it. modelcontextprotocol.io names Claude, ChatGPT, Visual Studio Code and Cursor among the clients that support the protocol.

Cleanlist does not publish a tested setup guide for ChatGPT, Cursor or Perplexity today, so treat those as the generic remote-MCP path rather than a supported one-click install. Competing servers publish wider matrices: ZoomInfo lists "Claude, ChatGPT, Perplexity, Replit, or custom AI agents" and Apollo lists Claude, ChatGPT and Perplexity (both fetched August 7, 2026). If cross-client coverage is a hard requirement, the Cleanlist REST API v2 works from anything that can send an HTTP request.

The Cleanlist MCP server works with any MCP client that supports remote servers over HTTP with OAuth, which includes ChatGPT, because mcp.cleanlist.ai speaks the standard protocol. Claude is the client Cleanlist supports and tests first, installed from Claude Settings, then Connectors, as a custom connector at https://mcp.cleanlist.ai/v1/mcp. Cleanlist publishes no tested ChatGPT setup guide, so treat that path as unverified rather than promised.

Cleanlist does not ship a first-party ChatGPT app or a Perplexity connector today, so in those clients you are on the generic remote-MCP path rather than a one-click install. If you need something versioned and stable outside Claude, the Cleanlist REST API v2 covers the same surface with a clapi_ Bearer key.

The Cleanlist MCP server exposes the tools listed in the table below, which is the live tool list rather than a count of the REST endpoints behind it. The groups are: people search, company search and find-similar-companies; person, company and bulk enrichment through the provider waterfall; smart agents that write AI research columns over a list; lead-list CRUD with CSV import; CRM sync, two way, to HubSpot, Salesforce and Pipedrive, plus one way out to the sequencers Outreach, Salesloft and Lemlist; CSV and JSON export; and account tools for identity, credit balance, credit estimates and usage.

Search and read tools are free to call. Enrichment, smart agents, saving leads and CRM sync are the tools that consume credits.

The Cleanlist tool names an agent sees in a connected session include search_people, search_companies, enrich_person, enrich_list, add_person_enrichments, add_company_enrichments, estimate_cost, create_list, get_lists, add_leads_to_list, list_leads_in_list, remove_leads_from_list, delete_list and whoami. Names map onto Cleanlist REST API v2 endpoints, so search_people is POST /search/people and estimate_cost is POST /credits/estimate.

Treat the list your own client shows as authoritative. Cleanlist ships 30+ tools across the full server and a given connector may expose a narrower curated set, with the remaining jobs (CRM and sequencer sync, smart agents, similar-company search, CSV file import) run in the Cleanlist app instead. Read tools/list before you write a prompt or a skill that assumes a tool exists.

The Cleanlist MCP server is free to connect and free to search. People search, company search, reading lists and checking your balance all cost 0 credits, on every plan including the $0 Free tier, which comes with 30 credits a month and no card. What limits searching is a daily quota of 60 People Searches per UTC day per key rather than your credit balance. An agent can explore an ICP, refine filters and size the market without spending anything.

Enrichment is where credits go, and Cleanlist bills for results rather than attempts: 1 credit for a verified email, 10 for a phone, 11 for both, and a lookup that returns nothing is not charged. Paid plans buy credits rather than access, at $79/mo for 1,500 (Starter), $229/mo for 5,000 (Pro) and $599/mo for 15,000 (Scale), with 25% off annual. Apollo takes a similar line on access, stating "Any Apollo plan qualifies, including free" (fetched August 7, 2026).

Cleanlist charges MCP tool calls in credits at the same rates as the app and the REST API: 1 credit for a verified email, 10 for a phone number, 11 for a full contact record, 1 for a company enrichment, 0.5 to save a newly added lead to a list, 0.2 per lead for CRM or sequencer sync, and 0.5 to 3 per lead for a smart agent depending on the agent type.

People search and company search cost 0 credits, as do reading lists and checking your balance. There is no separate MCP subscription. Credits come from one shared organization wallet, so an agent, a teammate working in the app and a script hitting the API all draw from the same pool.

Cleanlist publishes no fixed per-call contact ceiling for MCP. The pattern that scales is a list: create a lead list, add leads to it at 0.5 credits per newly saved lead, then call enrich_list against the list id with a signed quote, and poll until the run settles. One enrich_list call can cover a whole list, and bulk enrichment is asynchronous, so the size of the job is bounded by your credit balance rather than by a batch limit.

Throughput is governed by rate limits instead: 60 requests per minute per organization, 30 per minute per API key, and 60 People Searches per UTC day per key. Searches also default to 25 results, so an agent that wants volume has to ask for it explicitly. Cleanlist publishes no latency figure for enrichment, and inventing one would be worse than omitting it, so plan for a poll loop rather than a fixed wait.

Cleanlist enforces 60 requests per minute per organization, 30 requests per minute per API key, and a hard cap of 60 People Searches per UTC day per API key. Those limits sit on the REST API v2 endpoints, and MCP tool calls run against the same endpoints inside the same organization, so the per-organization ceiling and the daily search cap apply to agent traffic too. The 30-per-minute figure is scoped to a clapi_ API key rather than to an OAuth connector. There is a third tier the docs publish for REST traffic, 30 requests per minute for an organization on an AppSumo lifetime license without a paid plan (docs.cleanlist.ai/mcp-api/introduction), though AppSumo licenses are excluded from the MCP beta anyway. Exceeding a limit returns a 429 with a Retry-After header.

The daily search cap is the one that bites in agent workflows. An agent looping over many narrow searches will exhaust 60 in a single day, so batch filters into fewer, broader searches and page through the results. Cleanlist publishes no uptime SLA and the MCP server is in beta, so treat it as a working integration rather than a contractual one.

The Cleanlist MCP server is in beta on Cleanlist self-serve accounts, and AppSumo licenses are excluded from the beta. The gate is account type rather than plan tier, which is a different gate from programmatic REST API access. For the REST API the Cleanlist docs state it directly: "The public API requires a Pro plan or above (self-serve) or an AppSumo Tier 5+ license" (docs.cleanlist.ai/mcp-api/introduction, checked August 7, 2026). The docs do not publish a separate MCP gate, so the beta condition above is the one to go by.

Self-serve pricing is Free at $0 with 30 credits per month and no card, Starter at $79/mo with 1,500 credits, Pro at $229/mo with 5,000 credits, and Scale at $599/mo with 15,000 credits. Annual billing takes 25% off. Full detail is on the Cleanlist pricing page.

Server-side MCP means the MCP server runs remotely on the vendor infrastructure and the client connects to it over HTTP with OAuth. In-context MCP means the server runs locally beside the client, usually over stdio, with the process and credentials on your own machine and every tool definition loaded into the model context window.

Cleanlist is server-side only. https://mcp.cleanlist.ai/v1/mcp is a hosted remote endpoint, so there is nothing to install, credentials live in an OAuth grant instead of a local config file, and Cleanlist can add or fix tools without you upgrading anything. There is no published Cleanlist package on npm today, so the in-context path is not a supported install. The tradeoff you are accepting: server-side needs network access and an authorized connector, and every tool definition still loads into your context window at connect time, so keep the number of connected servers small if context budget matters.

Cleanlist returns entities and context, and it does not return events or signals. In the vocabulary agent builders use, an entity is a person or company record, context is the enriched detail an agent reads before it acts, an event is something that happened at a point in time, and a signal is an inference drawn from those events. Cleanlist MCP covers the first two: person and company entities from structured search, plus verified email, phone, title, seniority, location, education, past roles, industry and headcount as context on each record.

Cleanlist has no event feed, no intent data and no buying-signal product, and there are no revenue, funding-stage or technographic filters. Vendors that sell that layer say so on their own pages: one lists signals tracking among its 24 documented MCP tools (docs.lusha.com/mcp-docs, fetched August 7, 2026). If your agent needs intent or technographics, Cleanlist is not the server for that job.

No. Cleanlist gates every paid operation behind a signed quote. The agent must first call estimate_cost, which returns an HMAC-signed quote_id with a five-minute TTL plus your balance and whether it covers the job, and the server rejects the paid call if the real cost would exceed the quote or the quote has expired.

Three things make this hard to get wrong. Search, reading lists and checking your balance cost 0 credits, so an exploring agent burns nothing. Enrichment is pay-for-results, so a lookup that returns no email is not charged and an async run reserves credits up front and refunds the unused portion when it settles. Your balance is a hard ceiling: the wallet is shared across the app, the REST API and MCP, so the most an agent can consume is what you already hold.

Yes. Whatever a Cleanlist tool returns goes back into the model's context, because that is how MCP works: the client hands tool results to the model so it can use them in its next turn. Names, job titles, verified emails and phone numbers that Claude retrieves through Cleanlist are handled under your agreement with the model provider, not under Cleanlist's. A data-processing review therefore has to cover the MCP client and Cleanlist separately.

Cleanlist controls the other half of the path. The connector authorizes over OAuth scoped to your Cleanlist organization, so no raw API key sits inside a chat client, and revoking it from Claude Settings, then Connectors, cuts access immediately. Cleanlist publishes no SOC 2 report and no uptime SLA, so if your review needs formal attestations, wait for them rather than assuming them.

Cleanlist limits what an agent can touch in three ways: OAuth scoped to your Cleanlist organization, so you never paste a raw key into a chat client; a one-way sync tool, which pushes enriched leads out to HubSpot, Salesforce or Pipedrive, and to Outreach, Salesloft or Lemlist; and revocation, which you can trigger from Claude Settings, then Connectors, at any moment.

What Cleanlist does not claim: there is no published SOC 2 report, no uptime SLA, and no read-only key mode, because scopes are granted per user rather than per key. The MCP server is in beta. If your security review needs formal attestations, wait for them rather than assuming them.

The Cleanlist MCP server and Cleanlist REST API v2 cover the same surface and differ in who does the calling. MCP is for an AI client: Claude connects over OAuth, discovers 30+ tools, and decides which to call from your prompt. The REST API is for your code: you authenticate with a clapi_ Bearer key, call the roughly 30 endpoints yourself, and handle async workflow polling.

Credits, rate limits and the signed-quote model are identical. Two practical differences: REST API access requires a Pro plan or above while the MCP beta is gated by account type, and REST API v2 has no webhooks, so both surfaces poll for async enrichment results.

Yes, by delegation. Claude reads your plain-English request and calls the structured Cleanlist search_people tool with filters. The Cleanlist server takes filters, not prose. Ask for VPs of Sales at US SaaS companies with 201 to 500 employees and Claude maps that onto the title, seniority, country and headcount fields.

Cleanlist exposes about 24 distinct filter fields covering title, seniority, years in role, location, skills, education, past titles and companies, industry and headcount. There are no revenue, funding-stage, technographic or intent filters. The 2026 shorthand for steering a data tool this way is vibe prospecting: building a lead list by describing it conversationally instead of filling in a filter form. Anthropic's connector directory lists a vendor under that exact name at claude.com/connectors/vibeprospecting.