Skip to content
Kitbay
MCP serverexperiment

deep-research-mcp

Runs locallyMITUpdated on Jan 28, 2026325GitHub starsNeeds an API key

Open Deep Research MCP Server runs iterative web research on a topic, available as a local MCP tool or a command-line program.

What it does

Open Deep Research MCP Server runs iterative web research on a topic, available as a local MCP tool or a command-line program. It generates search queries, searches and scrapes pages through Firecrawl, which can be the hosted API or a local instance, scores each source's reliability between 0 and 1, and goes deeper or wider according to depth and breadth settings before writing a Markdown report with sources. Users can state source preferences in plain language, such as avoiding forums. Models from OpenAI, Anthropic, Google or xAI are chosen per run with the matching API key. The README calls the repository an agent-coding experiment mostly written by LLMs.

Capabilities

  • Run depth- and breadth-controlled web research
  • Score source reliability
  • Write a Markdown report with sources
  • Filter sources by plain-language preferences