OpenLeads

What OpenLeads MCP server does

OpenLeads MCP server connects an AI assistant to a local lead-research engine. It accepts natural-language searches such as dentists in Austin, fintech founders, or law firms in London, then selects relevant public sources and returns businesses, named people, and associated work email addresses.

The search system can use OpenStreetMap, Y Combinator, Hacker News, GitHub, OpenAlex, SEC EDGAR, Wikidata, company websites, and the NPI Registry. The sources depend on the query; users do not normally need to select one. A source can be pinned from the command line when needed.

Email results include evidence labels. A found address was published or confirmed by a mail server, a pattern address follows a format observed at the company, and a guessed address is an unconfirmed common format. These results are also grouped into safe, risky, or bad sending tiers.

How it works

The engine parses the request, federates the search across applicable public sources, and discovers people from company team and about pages. It applies a real-name check before creating a lead. Email checks can include published-address collection, company-pattern detection, MX, SPF, DMARC, Gravatar, and SMTP checks when port 25 is available.

OpenLeads MCP server runs on the user’s machine through the openleads mcp command. The README shows it being registered as a local command-based MCP server, so the assistant communicates with the local process rather than a hosted OpenLeads service.

The broader CLI also supports lead search, enrichment, single-address verification, drafting, sending, scheduling, workflows, and exports. Sending is a dry run by default; live sending requires an explicit option.

Setup and configuration

Install the package with pipx install openleads, pip install openleads, or npx openleads. The recommended extra is openleads[all]; the minimal Python installation provides a standard-library-only engine and CLI. Python 3.8 or newer is supported.

Register the MCP process with a client using the command claude mcp add openleads -- openleads mcp, or configure a server entry whose command is openleads and whose argument is mcp. The openleads doctor command checks local setup details including DNS, port 25, and mailbox configuration.

No API keys are required for the public-data search described in the README. The application can also start a local dashboard with openleads web, but the MCP setup itself uses the command-line process.

Tools and capabilities

OpenLeads MCP server is intended for conversational lead discovery and verification. Supported use includes:

• Finding people and companies from natural-language criteria.
• Returning work emails with evidence and delivery-risk labels.
• Searching public data across multiple source types.
• Checking an individual address through the verification workflow.
• Enriching a supplied list through the wider CLI.

The wider product can write email drafts, send messages, schedule recurring searches, monitor for new matches, and export leads to CSV, Google Sheets, Notion, Airtable, or a webhook. Those features are documented as CLI or local workflow functions; the excerpt does not provide separate MCP tool names for them.

Limitations and notes

Results come from free public data, so coverage and email certainty vary by query and source. An address marked guessed is not confirmed, while pattern evidence indicates a format rather than proof that the mailbox exists. Risk labels should be reviewed before outreach.

SMTP checks depend on whether port 25 is open, and the doctor command can identify related local configuration issues. The README reports benchmark results for selected searches, but those figures should not be treated as a guarantee for every location, industry, or person search.

The software runs locally and does not require API keys for its stated public-data searches. Sending is disabled by default through dry-run behavior, and live sending must be explicitly enabled.