# AI agents

> How larch builds AI agents: Scout’s agent runs a B2B sales pipeline end to end, and Claude Code works LinkedIn through Bridddge’s MCP server.

URL: https://larch.ee/technologies/ai-agents/

larch builds AI agents that do real work in shipped products. Scout’s agent runs a B2B sales pipeline from company discovery to drafted outreach, and in Bridddge, Claude Code works LinkedIn through a built-in MCP server.

- Products built: 2
- Lead score: 1 to 5
- Cycles per run: 10+
- Models: Claude · GPT
- Bridddge’s agent: Claude Code

## Built with AI agents

Two products, agents at work. One agent runs a B2B sales pipeline end to end. In the other product, Claude Code does LinkedIn work through tools we built.

### [Bridddge](https://larch.ee/work/bridddge/)

Bridddge’s MCP server gives Claude Code 30+ tools for LinkedIn: finding prospects, analyzing networks, writing outreach and managing campaigns.

- Year: 2025
- Stack: Electron, React, TypeScript, shadcn/ui, Tailwind CSS, SQLite, MCP, Claude AI
- Live product: https://bridddge.com

### [Scout](https://larch.ee/work/scout/)

The agent discovers companies, enriches contacts, scores each lead from 1 to 5 against the ideal customer profile and drafts personalized outreach.

- Year: 2025
- Stack: Next.js, React, TypeScript, Tailwind CSS, SQLite, Playwright, AI agents

## Related service

### [AI and automation](https://larch.ee/services/ai-automation/)

Useful intelligence, wired into the tools your team already works in.

What you get:

- **AI agents that work in production, not just demos**: Scout’s agent runs 10+ cycles on its own to reach a lead target.
- **MCP servers, LLM orchestration and multi-API pipelines**: Scout reaches Claude and GPT models through the OpenClaw gateway; Bridddge exposes its tools over MCP.
- **Scraping, data enrichment and automated daily workflows**: Scout finds founders and decision-makers on the web with Playwright and the Serper API.

## How we use AI agents

Agents that do the work. Sales prospecting is repetitive: finding companies, hunting for contacts, qualifying leads and writing cold emails. We built an agent to do it, and an interface to follow it.

### Scout: the whole pipeline

Scout’s agent is a modular pipeline with discrete steps: discover, enrich, qualify, expand and draft. Point it at an ideal customer profile and it works through those steps on its own, running 10+ cycles to hit a lead target.

Discovery searches configurable sources such as Product Hunt, Y Combinator, Crunchbase, GitHub, AngelList and Reddit communities. The agent deprioritizes the sources that perform badly and doubles down on the ones that work. Enrichment then finds the founders, CEOs and other decision-makers at each company, with their LinkedIn profiles and email addresses, using Playwright in stealth mode and the Serper API.

Every lead gets a score from 1 to 5 against the profile, based on industry, company stage, team size, budget, tech stack and founder type, and the agent explains its reasoning. The top-scoring leads seed a second wave of discovery: competitors, companies from the same accelerator cohorts and related players in niche verticals. Last, the agent drafts personalized cold emails from the user’s knowledge base of case studies, pricing documents and company profiles.

The models sit behind the OpenClaw gateway, which abstracts Claude and GPT so the provider can change. The user follows the agent in a Next.js app that updates live during a run, with a Kanban pipeline and an activity log of every agent action.

### Bridddge: Claude Code on LinkedIn

In Bridddge, the agent is Claude Code. Bridddge’s built-in MCP server gives it 30+ tools, so it can search for prospects, analyze networks, write personalized outreach and manage campaigns from plain-language requests. Bridddge’s own automation engine runs next to it, with human-like delays of 20 to 45 seconds and daily rate limits.

## Other technologies

Used alongside it. Bridddge’s agent works through an MCP server, and Scout’s runs behind a Next.js app. See what else we built with them.

- [MCP servers](https://larch.ee/technologies/mcp-servers/): seo-mcp brings SEO audits into AI editors, and Bridddge ships with its own MCP server.
- [Next.js](https://larch.ee/technologies/nextjs/): Hiki’s multilingual catalog front end and the Scout web app, both on Next.js 15 with the App Router.
- [All technologies](https://larch.ee/technologies/): Seven technologies, each one running in a product we shipped.

## How to start

Tell us what you want to ship. Bring the idea. We’ll tell you how we’d build it and what it takes.

- Book a 30-minute call: https://calendly.com/nicholas-stackwise/30min
- Email: hello@larch.ee