Coresignal MCP v2 is Coresignal's official Model Context Protocol server, with the goal of bringing fresh B2B data—895M+ employees, 70M+ companies, and 468M+ job posting records—directly into AI assistants. Users can use natural language in Claude, Cursor, Codex, VS Code, or any MCP-compatible client to search, pull full records, enrich contacts with verified email addresses, and export large result sets as downloadable files.
[Main capabilities]
- Natural language search: Retrieve employee, company, or job records in everyday language, for example "Senior Python developers at fintech companies in French". Each call costs a fixed 20 credits and returns total_count (the exact number of matches), up to 20 preview rows (limit=0 returns only the count), with each row showing the matched fields as evidence that the result appeared; it also returns a cache_id (a handle valid for 1 hour), which can be passed to entity_fetch to directly collect matching records without rerunning the search. Parsing the cache_id itself is free, and record IDs are kept on the server side.
- Field discovery: entity_fields supports free, instant lookup of field names by meaning, for example "salary" finds compensation fields, and "current job title and seniority" finds active_experience_title, experience.position_title, etc., without loading a vocabulary of 300+ fields into context.
- Record collection: entity_fetch can select up to 20 IDs from search results, or collect up to 1,000 records per call via cache_id; billing is based on records found, at 20 credits for employees/companies and 1 credit for jobs.
- Email enrichment: employee_enrich provides verified, deliverable business email addresses for up to 1,000 employee IDs, at 10 credits per email found, with IDs not found being free; EEA/UK contacts are unavailable due to GDPR.
- File download: Large results are not dumped directly into the conversation; instead, they are stored server-side as filesand a signed HTTPS download link valid for 1 hour is returned, to save LLM input tokens.
- artifact_read: Read back delivered files page by page through an authenticated MCP session, free of charge (records were already billed at fetch time), suitable for chat clients that cannot open links or access the file system.
[How to use] Requires a Coresignal account, an active subscription (credits), and a team API key configured in the console; after logging in, the MCP server automatically resolves the team key, so there is no need to paste it into a configuration file. On first connection, the client opens a browser window to log in to the Coresignal console.
- Claude Desktop: Natively supports remote MCP servers through Connectors. Open Settings → Connectors → Add custom connector, enter Coresignal as the name, and for the URL enter https://mcp.coresignal.com/mcp/v2，点击 After Add, Connect, open the browser to log in. Important: Claude Desktop will block downloads from unknown domains. If you need to download result files, add mcp.coresignal.com to the Domain allowlist in settings (managed by administrators for Team/Enterprise); otherwise the download link will be blocked, but you can still use the free artifact_read to read the data back in the conversation.
- Claude Code: run /mcp in the session, select coresignal, and complete the browser login; run /mcp again when re-authentication is needed.
- Cursor: add it to ~/.cursor/mcp.json; after Cursor detects that authentication is required, it will show a Needs login prompt. Click it to complete the browser login.
- VS Code: run MCP: Add Server from the command palette, select HTTP and enter the URL, or add it to .vscode/mcp.json; on first use it will prompt for authorization and handle OAuth in the browser.
- Cline: add it to ~/.cline/data/settings/cline_mcp_settings.json; open the Cline MCP interface, select Authorize OAuth from the menu and choose coresignal, and the browser will handle OAuth.
[Downloading and File Reading] Large results (and any call with delivery="url") are saved as files and return a signed download link valid for 1 hour. There are three ways for the agent to access the data: first, download the file yourself—in clients with a shell, this is usually done with curl to save it to disk, then use grep/jq/pandas for local analysis; second, read it back in the conversation—in clients without a shell and file system, call artifact_read(artifact_name, offset, limit) to read it in pages, which is free and universal and will not flood the conversation; third, manual download—if the agent is not allowed to obtain the link (the domain is not allowlisted, or the sandbox has no network), you can click the link to save the file and tell the agent its location. It is recommended to prioritize downloading: artifact_read is free on Coresignal credits, but not free on LLM tokens—each pagination will enterEnter the conversation and be re-billed by input tokens in subsequent turns, and each page is trimmed to a fixed token budget (a complete employee record is about 8k tokens); reading a large file may require hundreds of calls and exhaust the context window; whereas downloading a file consumes almost no tokens, and the assistant can use grep/jq to filter thousands of lines locally. Therefore, it is recommended to set up downloading in advance (domain allowlist, sandbox network access), and keep artifact_read only for clients that cannot download or for quickly viewing a small number of records. If downloading is blocked, the data is already stored and already paid for, so just use artifact_read to read it; do not rerun the fetch to "recover" the file, otherwise you will be billed twice.
[Billing and Pricing] Each response includes credits_consumed (the actual billed amount), for example, sending 1,000 IDs but being billed for 950 records, which can be used to determine the number of IDs not found. The server does not consume silently: before fetching, it first confirms the field scope with the user, and fetches that cannot be delivered will fail before consuming credits. Credits come from the team's Coresignal subscription, and keys and billing can be managed in the console. New users can use a 7-day free trial, which includes 2,000 credits, equivalent to 100 searches, or 100 employee/company records, or a mix; for a comparison of specific plans, see the official plan comparison page.
[Applicable Scenarios and Limitations] Suitable for B2B lead generation, in-depth company research, job and talent search, contact email enrichment, and large-scale data export in AI assistants or AI workflows. Example prompts include: find Nordic B2B SaaS companies that raised funding in the past two years and have 50–200 employees; search for data leaders at U.S. companies with 500+ employees and fetch 20 complete profiles along with verified emails to a file; pull the complete record for flo.health (funding rounds, employee growth, current job openings). Limitations and notes: a Coresignal subscription and credits are required; EEA/UK contact emails cannot be obtained (GDPR); the download feature depends on a domain allowlist or network access; although artifact_read does not consume Coresignal credits, it does consume LLM tokens, so it is not suitable for reading large files.