Announcing Autobound's MCP Server: B2B Signal Data for Claude Code, Cursor, and Every AI Agent

Most B2B data lives behind dashboards nobody opens.

Your team pays six figures a year for intent data, hiring signals, funding alerts. That data sits in a CSV export, a Slack notification channel, or a dashboard tab that someone checks twice a month. The data itself is often excellent. The access pattern is dead on arrival.

Here's what changed over the last 12 months: your team's AI tools now consume more structured data than your humans do. Claude Code pulls context while writing code. Cursor enriches prompts with live data. Custom agents make routing decisions based on account signals. ChatGPT research assistants synthesize company intelligence on demand.

None of these tools open dashboards. They call APIs. They query MCP servers. They pull structured data into context windows and reason over it.

We watched this shift happen across our customer base. The teams extracting the most value from Autobound's signal data stopped being the ones with polished BI dashboards. They're the ones whose AI agents have native, always-on access to live company intelligence.

So we built an MCP server and published it on npm.

What This Is

@autobound-ai/mcp-server is a single npm package. Install it, point it at your API key, configure your AI tool. Your Claude Desktop, Cursor, Claude Code, VS Code, or any custom MCP-compatible agent now has 11 tools that query Autobound's full signal data platform:

Search & Discovery

• searchCompanies — find companies by signal type, industry, employee count, recency window
• searchContacts — find people with active signals (job changes, LinkedIn posts, podcast appearances, conference talks)

Enrichment

• enrichCompany — full signal profile for a known domain. Every signal we've ever detected, structured.
• enrichContact — full activity feed for a known person. What they've posted, where they've spoken, when they changed jobs.

Signal Intelligence

• getSignalById — pull a single signal record by ID with full metadata
• listSignalTypes — browse every signal category, subcategory, refresh cadence, and delivery method
• signalTypeHistory — delivery timeline for a specific signal type (when did it last run, how many records)

Account Management

• getAccount — workspace info, plan, creation date
• getCredits — credit balance and usage
• getRequestLogs — recent API activity
• getStats — aggregate usage statistics

That's the entire surface. No SDK dependencies. No custom framework. No 47-page integration guide.

60-Second Setup

Install:

npm install @autobound-ai/mcp-server

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "autobound-signals": {
      "command": "npx",
      "args": ["@autobound-ai/mcp-server"],
      "env": {
        "AUTOBOUND_API_KEY": "ab_live_your_key_here"
      }
    }
  }
}

Restart Claude Desktop. That's it.

For Cursor, the config lives in .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "autobound-signals": {
      "command": "npx",
      "args": ["@autobound-ai/mcp-server"],
      "env": {
        "AUTOBOUND_API_KEY": "ab_live_your_key_here"
      }
    }
  }
}

For Claude Code (CLI):

claude mcp add autobound-signals -- npx @autobound-ai/mcp-server

Set AUTOBOUND_API_KEY in your environment. Done.

What People Actually Ask Their AI Tools

Once the MCP server is connected, here's what we see customers doing:

Sales reps in Claude Desktop:

• "What signals does Snowflake have in the last 30 days? Anything I can use for outreach?"
• "Find me 20 mid-market SaaS companies that filed SEC Form D funding this quarter"
• "Show me podcast appearances by C-level execs at companies with 200-1000 employees in fintech"
• "What's the latest news on Datadog? Any hiring velocity changes?"

Engineers in Cursor / Claude Code:

• "What does the hiring-trends signal record schema look like? Show me a real example."
• "Find companies with 5+ distinct signal types active in the last 7 days. I'm building a scoring model and need sample data."
• "What's the refresh cadence for each signal type? I need to know how often to poll."
• "Search for companies in my prospect list that have conference speaker signals — here are the domains: [list]"

RevOps teams running custom agents (LangChain, LlamaIndex, CrewAI):

• Funding signal detected → check if company is in CRM → if not, create lead → enrich with full signal profile → draft outreach → queue for review
• Monitor patent filings in specific technology categories → alert account owners when target accounts file
• Daily scan: pull all signals for accounts in "Evaluation" pipeline stage → prioritize by signal density → assign next actions
• Weekly: compare signal activity across competitor install base → surface accounts showing churn signals

Why MCP Matters More Than Another REST API

We already have a REST API. Every B2B data company does. The MCP server isn't a better API. It's a fundamentally different access pattern.

REST API = your engineers write integration code. They design a schema, build a pipeline, create endpoints in your product, build UI on top. That integration serves exactly one application. It takes 2-8 weeks depending on complexity. When you want another application to access the same data, you build another integration.

MCP Server = every MCP-compatible tool in your entire stack gets access simultaneously. Claude Desktop, Cursor, Claude Code, VS Code with Continue, ChatGPT with plugins, your custom LangChain agent, your team's internal research bot. All of them. One install, one API key.

The leverage is qualitatively different. You're not building "a better integration." You're turning your signal data into infrastructure that every AI tool can consume without per-tool integration work.

Think of it like this: the REST API is how your product talks to Autobound. The MCP server is how your entire team's AI layer talks to Autobound.

Multi-Tenant. Platform-Safe.

Every API key is scoped to a workspace. Your data, your credits, your rate limits. Completely isolated.

For platform builders: you can embed Autobound's MCP server into your own product. Your customers each get their own connection scoped to their own API key. Ship an "Add B2B Signals" button in your AI agent product that configures the MCP server for each customer.

We already have OEM partners doing this. If you're building GTM tools, AI SDR products, or research agents — the MCP server is how you give your product access to 35+ signal categories and 700+ signal subtypes without building signal infrastructure from scratch.

The Architecture

The MCP server runs as a local stdio process by default. Your AI tool (the "host") spawns it, connects via JSON-RPC 2.0 over stdin/stdout, discovers the available tools, and calls them as needed.

[Claude Desktop / Cursor / Custom Agent]
         |
         | JSON-RPC 2.0 (stdio)
         |
[@autobound-ai/mcp-server]
         |
         | HTTPS (Bearer auth)
         |
[Autobound Signal API]
         |
         | (Postgres, BigQuery, GCS)
         |
[35+ Signal Pipelines]

All computation happens server-side. The MCP server is a thin authenticated bridge between your AI tool and our API. No data is cached locally. No credentials are exposed to the AI model. The host application (Claude, Cursor) enforces tool approval policies — agents request permission before calling tools.

For remote deployment (shared team server, cloud hosting), the MCP server also supports HTTP+SSE transport. Deploy it once, connect multiple team members.

What Signal Data Is Available

Everything in the Autobound platform. 35+ signal categories, 700+ individual signal subtypes:

• Funding & Financial — SEC Form D private funding, 10-K/10-Q/8-K filings, earnings transcripts, financial reports
• Hiring & Workforce — hiring trends by role/seniority/department, hiring velocity, employee growth, job changes
• News & Media — 40 news subtypes (M&A, partnerships, funding, product launches, leadership changes, bankruptcies...), daily delivery
• Social & Content — LinkedIn posts (company + contact), Twitter, Reddit, YouTube, GitHub activity
• Executive Intelligence — podcast appearances with full transcripts, conference speaker commitments, work milestones
• Product & Technology — Product Hunt launches, patent filings, Hacker News mentions, product reviews, website intelligence
• Market & Competition — SEO traffic analysis, Glassdoor reviews, federal contract awards
• Buyer Intent — topic-level research activity across the B2B web (38K+ topics)

Each signal type has a defined refresh cadence (daily, weekly, monthly), a structured schema, and entity-level tracking so nothing gets dropped.

The Bigger Picture

18 months from now, most B2B data will be consumed by AI agents, not humans clicking through dashboards. Not because humans stop needing intelligence — because the interaction pattern shifts. You don't log into a tool and run a query. You ask your agent a question and it pulls what's relevant, synthesizes it, and surfaces the answer (or takes action autonomously).

Autobound's signal data ships three ways:

1. GCS / S3 flat files — bulk delivery into your data warehouse. Full datasets, your infrastructure.

2. REST API — your application queries us programmatically. Build products on top.

3. MCP Server — every AI tool in your stack, immediately. No integration per tool.

For teams already running Claude Code, Cursor, or building autonomous GTM workflows, delivery method #3 is becoming delivery method #1. It's the lowest-friction path from "we have signal data" to "our team actually uses signal data every day."

Get Started

Existing customers: Your current API key works. Install the package, add the config, restart your tool. You're live in 60 seconds.

New here:

npm install @autobound-ai/mcp-server

Get an API key at autobound.ai/developers. Self-service access opening later this quarter.

Resources:

• MCP Server docs →
• npm package →
• Full Signal API reference →
• Signal Directory (browse all 700+ subtypes) →
• Book a demo →