Projects01 AI / Agents
AI Research and Web Intelligence Agent
An agentic research workflow that crawls the web, calls tools over MCP, and returns a sourced brief through OpenRouter.
Problem
Researching a topic still means opening dozens of pages and losing the thread of what was already read, compared, or discarded.
Solution
Built an n8n agent that uses Firecrawl for pages, OpenRouter for reasoning, and MCP so the agent can fetch, compare, and summarize through tools instead of a hardcoded script for every site.
Architecture
- Research query
- n8n agent
- Firecrawl
- OpenRouter
- MCP tools
- Sourced brief
Technology
- n8n
- Firecrawl
- OpenRouter
- AI Agent
- MCP
Engineering decisions
- Tool calling that stays on the question instead of wandering the web.
- Crawled text that is clean enough to cite, not a blob of navigation chrome.
- MCP access that is narrow enough to be safe and wide enough to be useful.
- A brief that shows sources so a reader can check the agent's work.
Automation
Crawl, tool calls, and the first brief run unattended. A person still decides whether the brief is good enough to use.
Reliability
If a crawl or tool fails, the brief says what is missing instead of filling the gap with an unsourced claim.
Security
The OpenRouter key stays server-side. MCP tools expose only the fetches the workflow is allowed to make.
AI layer
The agent picks tools through MCP. OpenRouter runs the model. Claims in the brief have to point at a crawled page or a tool result.