Building an Enrichment Agent on the PredictLeads MCP Server

This is a hands-on walkthrough for building a specific, narrow agent: one whose only job is to take a company (by domain or name) and return a structured enrichment summary – technology stack, open roles, recent news, and funding activity – pulled live from the PredictLeads MCP Server.

If you’ve already read our guide on building a GTM agent on the PredictLeads MCP Server, this is the narrower, enrichment-specific version of that same idea.

What We’re Building

An agent that, given a company domain, returns a short brief covering:

• Current technology stack and recent technology changes.
• Open job postings, as a proxy for where the company is investing.
• Recent company news – launches, partnerships, leadership changes.
• Financing events, if any, and how recent they are.

The point isn’t a dashboard – it’s a single, on-demand answer a rep, analyst, or downstream workflow can request whenever they need it, instead of pulling it from a field that was last refreshed weeks ago.

Step 1: Connect the MCP Server

Add the PredictLeads MCP server to your agent client, pointing at https://mcp.predictleads.com/. For a quick setup, authenticate with HTTP headers using your API key and token, available from your Subscription Plans page:

• X-Api-Key: {your_api_key}
• X-Api-Token: {your_api_token}

For a production deployment, use the OAuth2 Client Credentials flow instead – Access Token URL https://oauth.predictleads.com/token, with your API key as the Client ID and your API token as the Client Secret. Exactly how you register the server depends on your agent framework; see the MCP integration docs for client-specific setup notes.

Step 2: Give the Agent a Narrow Job

Rather than exposing every PredictLeads tool with a generic “answer anything about this company” prompt, scope the agent’s instructions specifically: given a company domain, call the technology, job openings, news events, and financing events tools for that domain, and return a short structured summary combining all four. A narrow, well-scoped prompt produces more consistent output than an open-ended one, and it makes the agent’s behavior predictable enough to plug into a larger workflow.

Step 3: Example Run

Given a prompt like “Enrich acme-example.com,” the agent should:

• Call the technologies tool for the domain and summarize the current stack.
• Call the job openings tool and note roles that signal investment areas – engineering growth, new go-to-market hires, and so on.
• Call the news events tool and surface anything from the last 90 days.
• Call the financing events tool and note the most recent round, if any.
• Combine all four into a short brief, rather than four separate answers.

Because every call happens live, the brief reflects the company as it looks right now, not as it looked whenever a batch job last ran.

Step 4: Wiring It Into a Larger Workflow

Once the core enrichment agent works reliably, it’s straightforward to trigger it from other systems – a new-lead webhook, a Slack command a rep runs before a call, or a scheduled review of accounts that haven’t been looked at in a while. The agent itself doesn’t change; only what triggers it does.

A Note on Request Limits

Each account has a monthly request limit tied to its subscription plan. If this agent gets triggered automatically at volume – for example, on every inbound lead – check your plan’s limit first. Exceeding it returns a 402 HTTP error until the next billing cycle.

FAQ

Related Guides

• PredictLeads MCP Integration Docs
• Building a GTM Agent on the PredictLeads MCP Server
• PredictLeads MCP Integration Guide