CatchIntent LogoCatchIntent
Blog

We Made CatchIntent AI-Native — Here's What Changed

CatchIntent now supports MCP, letting users connect Claude, Cursor, and other AI tools directly to their intent data. Here's why we did it and what it means for the future of social listening.

Published on

Reading time

4 min read
We Made CatchIntent AI-Native — Here's What Changed

Last week we shipped something that changes how CatchIntent works fundamentally.

You can now connect Claude, Cursor, Windsurf, ChatGPT, or any MCP-compatible AI tool directly to your CatchIntent workspace. Search signals, manage listeners, draft outreach, enrich prospects, push to CRM — all through natural language.

This isn’t a chatbot bolted onto a dashboard. It’s a new interface to your data.

TL;DR: CatchIntent now runs as an MCP server. One command to connect, OAuth to authorize, 20 tools available. Works with all major AI tools. Included on every plan. We also shipped 6 pre-built GTM skills and a new Settings > Agents tab. Set it up →

Why We Did This

We’ve been watching the AI-native wave closely. Positz went from 20K to 70K MRR by going agentic for social media scheduling. The pattern is clear: users have AI access (Claude, Codex, Cursor). What they don’t have is data.

CatchIntent sits on a firehose of intent data — buying signals, prospect profiles, company intelligence, keyword performance metrics. That data was locked behind our dashboard.

MCP unlocks it.

What Changed

The Dashboard Stays

Nothing was removed. The dashboard is still the best way to onboard, manage your team, configure alerts, and get visual overviews. If you never touch MCP, everything works exactly as before.

MCP Adds a New Interface

Connect your AI tool and you get 20 tools:

  • Signal tools — search, filter, view details, update status, generate outreach
  • Listener tools — create, edit, toggle, review health
  • People tools — search, enrich, update pipeline, export CSV
  • CRM tools — push to HubSpot, Close, Pipedrive
  • Workspace tools — stats and usage

Every tool respects your plan limits and role permissions. Members can read, Admins can write. Same rules as the dashboard.

OAuth, Not API Keys

We didn’t go the traditional API key route. Instead, we implemented full OAuth 2.1 with PKCE. You run claude mcp add catchintent, your browser opens, you authorize, and you’re connected. Tokens refresh automatically. Revoke anytime from Settings > Agents.

This means any MCP client works out of the box — no copying API keys around.

Skills for GTM Teams

Raw tools are powerful but require knowing what to ask. So we built Skills — pre-built playbooks that tell your AI how to run common GTM workflows:

Terminal window
npx skills add catchintent/skills

Six workflows: morning signal review, outreach drafting, prospect enrichment, pipeline review, listener optimization, and weekly reports. Each skill combines multiple tools into a single natural language workflow.

What We Learned Building This

MCP OAuth is underspecified. The spec says “OAuth 2.1” but every client sends slightly different parameters. We ended up skipping strict DTO validation on OAuth endpoints entirely — just extract what we need and ignore the rest. This is the right approach for any MCP server.

Stateless mode is simpler. We create a fresh McpServer instance per request. No session state, no cleanup. It’s more CPU per request (tool registration happens every time) but operationally simpler and scales horizontally without session affinity.

Role-based access matters. An early version let any authenticated user do anything. We quickly added role checks — Members can read, Admins can write. If someone has Member access in the dashboard, they should have Member access through MCP. Parity matters.

Rate limiting is essential. Without it, a runaway AI agent could hammer your database. We implemented per-workspace sliding window rate limits in Redis, gated by plan tier.

What’s Next

This is the foundation. We’re exploring:

  • Agentic search via MCP — letting users run LinkedIn prospect searches from their AI tools using CatchIntent’s search infrastructure
  • Enterprise raw data access — for teams that want to run their own AI analysis on our ingested posts, bypassing our pipeline
  • MCP registry listings — getting CatchIntent into Claude’s connector directory, mcp.so, Smithery, and other discovery platforms

Try It

If you’re already a CatchIntent user:

Terminal window
claude mcp add catchintent --transport http https://engine.catchintent.com/mcp

If you’re new, sign up for a free trial. MCP access is included on all plans.

We’d love to hear what workflows you build with it. Tag us on X or LinkedIn.

Ready to catch buyer intent signals?

Start your 7-day free trial and discover high-intent leads from social conversations.

Find Your Buyers
Back to Blog