Key takeaways
- LangChain's configurable deep research agent — works across any LLM provider, search tool, and MCP server. Built on LangGraph for structured agent workflows
- Ranked #6 on the Deep Research Bench Leaderboard with a RACE score of 0.4344 — though the leaderboard space has since gone offline, so the ranking is historical
- 11.7k stars (up from 10.8k in March 2026), MIT license. Still maintained as of June 2026, but recent commits are dependency bumps rather than new features
- The "batteries included" approach — multi-provider support, MCP integration, and LangGraph orchestration out of the box. Free LangChain Academy course
FAQ
What is open_deep_research?
LangChain's open-source deep research agent built on LangGraph. It performs iterative web research using any LLM provider and search tool, with MCP server support and a no-code UI via LangGraph Studio.
Is open_deep_research still maintained?
Yes, but lightly. As of June 2026 the repo is active (last push June 7, 2026, not archived), though recent commits are dependency updates rather than feature work. The architecture has been stable since the July 2025 supervisor rewrite.
How does it compare to dzhng/deep-research?
dzhng is minimal (~500 LoC, zero dependencies). open_deep_research is feature-rich (LangGraph, multi-provider, MCP, UI). Trade-off between simplicity and capability.
Overview
Open Deep Research is LangChain's fully open-source deep research agent — a configurable system that works across any LLM provider, search tool, and MCP server. Built on LangGraph, it uses a three-phase architecture (scope, research, write) with a supervisor agent that delegates to parallel research sub-agents. It ranked #6 on the Deep Research Bench Leaderboard with a RACE score of 0.4344, on par with popular proprietary deep research products — though that leaderboard has since gone offline, so the ranking reflects a 2025 snapshot.
The project emphasizes configurability over minimalism — where dzhng/deep-research is ~500 lines, open_deep_research provides multi-provider support, MCP integration, LangGraph Studio UI, and a free course on building deep research agents from scratch.
Key stats (as of June 2026): 11,671 stars (up from 10,828 in March 2026), 1,664 forks, MIT license, Python. Last push June 7, 2026; not archived. Free LangChain Academy course available.
Key Features
- Multi-provider: Works with OpenAI, Anthropic, Google, and other LLM providers
- MCP support: Connect to MCP servers for additional tools and data sources
- LangGraph orchestration: Supervisor architecture coordinates parallel research sub-agents with context engineering to limit token bloat
- No-code UI: LangGraph Studio provides a visual interface for non-developers
- Benchmarked: Tested on Deep Research Bench with published scores (leaderboard now archived)
Pricing
Free and open source under the MIT license. You bring your own API keys — costs are whatever your chosen LLM provider and search API charge per run. The companion LangChain Academy course is also free; LangSmith tracing and LangGraph Platform hosting are optional paid LangChain services, not requirements.
Competitive Position
Strengths: Most configurable open-source deep research agent. LangChain ecosystem integration. MCP support. Benchmarked performance. Free course.
Weaknesses: LangChain/LangGraph dependency adds complexity. Heavier setup than minimal alternatives. Prompt-based (no fine-tuned model).
Cautions
- Feature development has slowed — as of June 2026, recent commits (May–June 2026) are dependency bumps, not new capabilities; the core architecture hasn't changed since the July 2025 supervisor rewrite
- No versioned releases — the GitHub repo publishes no tagged releases, so pinning a stable version means pinning a commit
- Benchmark evidence is stale — the Deep Research Bench Leaderboard space returns an error as of June 2026; the #6/0.4344 result survives only via the Wayback Machine
- Reference implementation, not a product — LangChain positions it as a starting point to fork and customize, not a supported tool
What Developers Say
As of June 2026, I could not find verbatim, attributable developer testimonials specifically about LangChain's open_deep_research — Hacker News submissions about it (including a November 2025 internals walkthrough) drew no substantive comment threads, and the deep-research discussions that do exist on HN center on rival implementations (HuggingFace's, btahir's, Zilliz's DeepSearcher). Community coverage is mostly tutorial-style blog posts rather than first-person usage reports. The 11.7k stars and 1,664 forks are the strongest available adoption signal.
Bottom Line
Recommended for: Teams already on LangChain/LangGraph who want a benchmarked, fork-and-customize deep research starting point with MCP and multi-provider support.
Not recommended for: Anyone wanting a minimal codebase (use dzhng/deep-research) or a maintained product with versioned releases and support guarantees.
Outlook: Stars keep climbing (10.8k → 11.7k since March 2026) and the repo isn't abandoned, but activity has settled into maintenance mode — dependency bumps, no new features, no releases. Treat it as a stable, well-documented reference architecture rather than an evolving project.
Research by Ry Walker Research