An AI agent is only as useful as the data it can reach. SignalHire built an MCP server so ChatGPT, Cursor, Windsurf, and any MCP-compatible client can query 850M+ professional profiles, pull verified contact details, and act on live company data, all inside the same chat window the agent already runs in.
This article covers why the server exists, how it works under the hood, what tools it exposes, how authentication works, three practical use cases, a step-by-step setup guide, and how pricing and usage limits apply.
Why We Built an MCP Server for SignalHire
Recruiters and sales teams lose time on a repeating loop. They search a database, copy a name, paste it into a spreadsheet, switch to an email finder, copy the result again, then paste it into a CRM. Each switch breaks focus and adds a chance for error.
The Model Context Protocol removes that loop. MCP is an open standard, comparable to USB-C for AI, that lets an assistant connect securely to outside tools and act on live data without a custom integration for every platform. Anthropic released MCP in November 2024, and within a year the protocol reached industry-wide adoption, with OpenAI, Google, Microsoft, and AWS all building support for it and the Linux Foundation now stewarding its governance.
SignalHire’s own contact database supplies the raw material an AI agent needs: 850M+ verified profiles, 30M+ companies, and 40+ data sources refreshed every seven to ten days. The MCP server connects that database directly to the chat window, so a recruiter or seller never leaves the conversation to find a name, enrich a contact, or check a headcount.
The market pressure behind this move is real.
- McKinsey’s State of AI report found that 62 percent of organizations are experimenting with or piloting AI agents, though fewer than 10 percent report scaling agents in any single business function.
- Gartner projects that 40 percent of enterprise applications will embed task-specific AI agents by the end of 2026, up from under 5 percent in 2025. An agent without a live data connection cannot clear that bar.
SignalHire’s Email Finder API closes that gap for anyone building or using an agent today.
How the SignalHire MCP Server Works
The SignalHire MCP server runs as a remote, hosted endpoint. Nothing installs locally, and nothing runs on the user’s machine. The client sends a request, the server checks a credential, and SignalHire’s database returns a result, formatted for the AI host to read and act on.
The architecture follows three layers. The AI host, meaning ChatGPT, or another assistant, sits at the front. The MCP client acts as a protocol bridge in the middle. SignalHire’s MCP server sits at the back, holding the data and executing the query. A prompt like “find 15 IT directors at SaaS companies with 200 to 1,000 employees” travels through that chain, and the agent decides which SignalHire tool answers it.
Connection takes one endpoint: https://mcp.signalhire.services/mcp, reached over MCP Streamable HTTP transport. The server holds no global key of its own. It reads whichever credential the request carries, and it detects the token type automatically, whether that is an OAuth token or a bearer API key.
Setup takes roughly 30 seconds for a non-developer using the ChatGPT connector UI, and it takes one CLI command for a developer using ChatGPT. Full technical detail on endpoints, transport, and tool schemas lives in the SignalHire API documentation.
What Tools the Server Exposes
An AI agent connected to SignalHire gains access to eleven tools, grouped into four categories, and the agent chooses which tool to call based on the prompt it receives.
Search and discovery tools let the agent query the database with filters: title, location, company, industry, seniority level, years of experience, and Boolean keyword logic. A pagination tool, scroll_search, fetches the next batch of results, and it must run within 15 seconds of the initial search call.
Profile retrieval tools separate two actions on purpose. retrieve_person_no_contacts pulls a candidate’s experience, skills, and education without spending a contact credit. retrieve_person_profile pulls the same profile plus verified emails, phone numbers, and social links, and this call spends one contact credit per successful match. That split lets an agent browse freely and reveal contact data only when a human confirms the spend.
Company tools resolve a name, domain, or slug into a stable numeric ID for free, then pull full company data, including exact headcount and headquarters location, for one company credit. Utility tools check remaining credits, check the daily search quota, and list supported seniority levels, departments, and all 146 industry categories, all at no cost.
Every tool accepts a response_format parameter, so an agent can request markdown for a chat reply, or CSV or JSON for a script. The remaining credit balance appends automatically to each response, so the agent, and the person reading its output, always sees the balance without a separate check.
Authentication and API Keys
SignalHire supports two authentication paths, and the right one depends on the client.
The distinction matters for a simple reason. OAuth suits a person clicking through a settings screen once. An API key suits a script, a CI pipeline, or an IDE that runs unattended and needs a credential it can store in a config file. SignalHire’s paid plans all include MCP access within the plan’s credit limits, so the authentication choice is about client type, not plan tier. Search and profile browsing without contact details work inside those limits too, but revealing a verified email or phone number draws on the plan’s contact credits, so that step requires an active paid plan.
The distinction matters for a simple reason. OAuth suits a person clicking through a settings screen once. An API key suits a script, a CI pipeline, or an IDE that runs unattended and needs a credential it can store in a config file. SignalHire’s plans, including the free tier, all include MCP access within the plan’s credit limits, so the authentication choice is about client type, not plan tier.
Key Use Cases for MCP
Three patterns cover most of what teams build once the SignalHire MCP server is connected. Each shifts a task that used to require three or four tools into a single conversation.
#1. Automate Lead Enrichment Inside an AI Agent
A sales or recruiting AI agent can search, filter, and enrich a list without a human touching a spreadsheet. The agent runs search_candidates_query against a filter set, such as director-level IT titles at companies of a given size, then calls retrieve_person_profile on each match to reveal verified emails and phone numbers.
Enrichment only spends a credit on a successful match. A failed lookup costs nothing, which removes the wasted spend that used to come from list-cleaning tools charging per attempt rather than per result. IBM estimates that bad B2B contact data costs US businesses roughly $3.1 trillion annually, a cost that traces directly to stale emails, duplicate records, and unverified phone numbers sitting in a CRM. An agent that enriches only on confirmed matches attacks that cost at the source.
From there, the agent can push the enriched list straight into a named list, a job project, or a CRM like HubSpot or Salesforce, no export file and no manual upload required.
#2. Ask ChatGPT to Find and Verify Emails
A recruiter or seller can connect SignalHire from ChatGPT’s app directory, then ask directly: “find the CMO’s email at this company, and verify it before you show me.” The agent calls search_candidates_query to locate the person, then retrieve_person_profile to reveal and verify the contact, and the response includes the credit cost so the user sees exactly what that answer spent.
This pattern replaces the old four-method workaround: guessing a pattern, scraping LinkedIn, checking a company’s press page, and running the result through a separate verifier. SignalHire’s bulk email finder and phone number finder tools sit behind the same MCP endpoint, so the same conversation that finds one contact can scale to fifty without switching context. Claude can run the same pattern too, through a custom connector or a JSON configuration, though SignalHire is not yet listed in a public connector marketplace for Claude the way it is in ChatGPT’s app directory. Either way, the agent decides which tool the prompt calls for, and the person only has to state the goal.
#3. Build Custom Workflows
A developer using ChatGPT can wire SignalHire into a larger pipeline: a nightly script that searches for new hires at target accounts, enriches the ones matching a title filter, and writes the result to an internal database or a Slack digest. The CLI connection, claude mcp add –transport http SignalHire …, takes one command, and from there the agent has the same eleven tools available inside a scripted, repeatable workflow rather than a one-off chat.
This path also suits internal tools built for a team rather than a single user. A recruiting ops team can build a ChatGPT agent that runs a saved search every Monday morning, checks the daily quota before running, and reports headcount changes at watched accounts, all without a human opening SignalHire’s own interface.
Pricing and Rate Limits
MCP access ships on every paid SignalHire plan, with plans starting at $49 per month. There is no separate MCP fee and no separate signup. The plan a team already holds determines its credit balance, and the MCP server draws from that same balance. Search and no-contact profile browsing run within those same limits, but revealing a verified email or phone number spends a contact credit, so that step requires an active paid plan.
Usage runs on two meters rather than a fixed requests-per-minute ceiling. Search and profile-browsing calls draw against a daily search quota, checked for free at any point with check_without_contacts_credits. Contact reveals draw against a contact credit balance, one credit per successful match, checked for free with check_balance. Company lookups draw against a separate company credit balance. Utility calls, such as listing supported industries or departments, cost nothing regardless of volume.
The credit-on-success model is the detail worth acting on. A team should budget for successful matches, not for attempts, since a failed enrichment call never spends a credit. That changes how a team should size a plan: watch match rate, not call volume, when deciding whether to upgrade.
What This Means Going Forward
The SignalHire MCP server turns a contact database into something an AI agent can act on directly, not something a person has to operate on the agent’s behalf. That distinction is where the real value sits. Gartner’s own data shows the gap between agent ambition and agent execution is still wide, with only 17 percent of organizations having deployed an AI agent to date against more than 60 percent that expect to within two years. A live, credit-metered data connection is one of the pieces that closes that gap, and it is available today, on the free tier, in about 30 seconds.
Possible next steps: install or connect the MCP server, read the MCP documentation, generate an API key, run a sample query inside ChatGPT, or book a demo with the SignalHire team. A live, credit-metered data connection is one of the pieces that closes that gap, and it is available today, on any paid SignalHire plan, in about 30 seconds.
FAQs
Which plans include MCP access?
Every paid SignalHire plan includes MCP access. A team connects today and runs the MCP tools inside its plan’s existing credit limits. There is no separate MCP fee and no separate signup. Search and no-contact profile browsing work inside that access, but revealing a verified email or phone number spends a contact credit, so that step requires an active paid plan.
Which AI tools does the SignalHire MCP server support?
ChatGPT connects through SignalHire’s listing in its app directory. Claude reaches the same MCP endpoint through a custom connector or a JSON configuration, though SignalHire is not yet listed in a public connector marketplace for Claude. Cursor and JetBrains IDEs connect the same way, and any MCP-compatible assistant that can reach a remote MCP server works too. The agent, not the person, decides which SignalHire tool a given prompt calls.
Is the SignalHire MCP server local or remote?
The server runs remotely, hosted at https://mcp.signalhire.services/mcp. Nothing installs locally and nothing runs on a user’s machine. A Claude.ai or Claude Desktop client reaches that hosted endpoint through OAuth in about 30 seconds.
How are credits charged on the MCP server?
Searches draw against a daily search quota tied to the plan. Enrichment, meaning email and phone appending, only charges a credit on a successful match, so a failed lookup costs nothing. That credit-on-success rule is the same one that governs every tool described earlier in this guide.
Do I need to be a developer to connect it?
No. Setup takes about 30 seconds and asks for one click: Authorize, inside the AI tool’s connector settings. No code and no API key entry are required for that path. A developer who wants the CLI or programmatic route still has it, through Claude Code, Cursor, or the Python SDK, but it is a choice, not a requirement.
