Awesome Research Agents

🔍 Awesome Research Agents

Curated resources for building, evaluating, and stress-testing research agents around Claude Code, OpenAI Codex CLI, and adjacent coding-agent CLIs — across AI/ML research workflows: literature, hypothesis design, implementation, experiments, analysis, writing, review, rebuttal, and project memory.

New here? Read [Start Here](#start-here). Shipping a workflow? Skip to [Coding-Agent Substrates and Hosts](#coding-agent-substrates-and-hosts) and [Skills, Workflow Packs, and Reusable Procedures](#skills-workflow-packs-and-reusable-procedures).

> If you maintain a research-agent project that should be on this list, or spot a mistake, broken link, or stale entry, please open an issue or pull request.

Contents

• [Start Here](#start-here)
• [Research Stage Map](#research-stage-map)
• [Coding-Agent Substrates and Hosts](#coding-agent-substrates-and-hosts)
• [Skills, Workflow Packs, and Reusable Procedures](#skills-workflow-packs-and-reusable-procedures)
• [Literature Discovery and Evidence Grounding](#literature-discovery-and-evidence-grounding)
• [Lab Integrations, Notebooks, Experiment Tracking, and Memory](#lab-integrations-notebooks-experiment-tracking-and-memory)
• [Autonomous Execution Loops](#autonomous-execution-loops)
• [End-to-End Research and Hypothesis Systems](#end-to-end-research-and-hypothesis-systems)
• [Manuscript, Figure, Reproduction, and Paper-to-Agent Systems](#manuscript-figure-reproduction-and-paper-to-agent-systems)
• [Review and Rebuttal Systems](#review-and-rebuttal-systems)
• [Benchmarks and Research Arenas](#benchmarks-and-research-arenas)
• [Practitioner Reports, Methodology Essays, and Agent Frameworks](#practitioner-reports-methodology-essays-and-agent-frameworks)
• [Open Problems and Gaps](#open-problems-and-gaps)
• [Contributing](#contributing)
• [Citation](#citation)
• [License](#license)

Tagging

Tags are conservative and used for within-section ordering (CLI-native → CLI-compatible → adjacent):

• CLI-native — ships explicit Claude Code/Codex skills, configs, docs, or workflows.
• CLI-compatible — usable from these CLIs but not primarily built for them.
• adjacent — informs research-agent design without being a CLI workflow.

Start Here

1. Claude Code — core Claude-native substrate: repo editing, shell, skills, hooks, subagents, memory, MCP.
2. OpenAI Codex CLI / Codex — OpenAI-native substrate: local CLI, AGENTS.md, skills, MCP, cloud/background tasks, parallel work.
3. AI-Research-SKILLs — broadest reusable research-skill library before writing your own.
4. ARIS — strongest Claude/Codex-facing autonomous research loop with claim-audit patterns.
5. PaperOrchestra — main paper-writing and figure-generation reference.
6. PaperQA2 — literature-grounded retrieval instead of model memory for citations.
7. ResearchArena (CLI-agent arena) — calibrates expectations on direct CLI research-agent evaluation.
8. Stop Automating Peer Review Without Rigorous Evaluation — read before deploying review/rebuttal agents.
9. AI for Auto-Research: Roadmap & User Guide — comprehensive lifecycle survey of the whole field as one map.

Research Stage Map

| Stage | Useful entries | |---|---| | Idea & literature | AI-Research-SKILLs, ARIS, PaperQA2, LitLLM, OpenScholar, OpenNovelty, STORM, ResearchAgent, ai-skill-scholar, paper-search-mcp | | Experiment design | ARIS, The Agentic Researcher, Agent Laboratory, AI co-scientist, EvoScientist, AAAR-1.0 | | Coding | Claude Code, OpenAI Codex CLI, Aider, Gemini CLI, OpenHands, PaperCoder / Paper2Code | | Execution | ARIS, W&B MCP Server, MLflow MCP Server, MLE-bench, MLAgentBench, CORE-Bench, RE-Bench, tinyKaggleClaw | | Analysis | Jupyter MCP Server, W&B Skills, MLflow MCP Server, data-to-paper, PaperOrchestra | | Writing | PaperOrchestra, Agent Laboratory, The AI Scientist, data-to-paper, GPT Researcher | | Peer review | ResearchArena, Stanford Agentic Reviewer, AgentReview, The AI Scientist, LLM-REVal, Re2 | | Rebuttal | Paper2Rebuttal, DRPG, ARIS, claude-code-my-workflow, ICML 2026 Peer Review FAQ | | Project memory | Claude Code, OpenAI Codex CLI, Letta Code, Obsidian Memory, Zotero MCP / ZotPilot, PARNESS, AgentRxiv |

Coding-Agent Substrates and Hosts

• Claude Code (2025, [CLI-native]) — Terminal-native coding agent that reads repos, edits files, runs commands, and composes with CLAUDE.md, skills, subagents, hooks, permissions, MCP, and SDK workflows.
• OpenAI Codex CLI / Codex (2025, [CLI-native]) — Local and cloud/background coding agent with AGENTS.md, skills, MCP, subagents, worktrees, and parallel task support for implementation, evals, analysis, and reports.
• OpenAI Codex cloud/background tasks (2026, [CLI-native]) — Codex web/cloud mode for delegating tasks into isolated GitHub-connected environments; maps to ablations, eval additions, and report-generation threads.
• Aider (2023, [CLI-compatible]) — Terminal pair-programming agent with repo maps, file editing, lint/test loops, broad model support, and automatic git commits.
• Gemini CLI (2025, [CLI-compatible]) — Open-source terminal agent with file tools, shell, web fetch, Google Search grounding, MCP, and GEMINI.md customization.
• OpenHands (2024, [CLI-compatible]) — Model-agnostic software-agent platform with SDK, CLI, local GUI, cloud/enterprise modes, and benchmark infrastructure.
• goose (2025, [CLI-compatible]) — Provider-agnostic agent for code, workflows, research, writing, automation, and data analysis with MCP extensions, recipes, subagents, and CLI/desktop/API modes.
• Cursor (2023, [adjacent]) — IDE-oriented coding agent frequently mentioned as a compatible host for skill packs and MCP workflows.

Skills, Workflow Packs, and Reusable Procedures

• Anthropic Agent Skills and Claude Code skills (2025, [CLI-native]) — Skills as progressively loaded folders with SKILL.md, scripts, resources, and optional dynamic context; native packaging pattern.
• OpenAI Codex Skills and plugins (2026, [CLI-native]) — Codex skills with progressive disclosure, skill locations, and plugin packaging for skills, app mappings, and MCP config.
• AI-Research-SKILLs (2026, [CLI-native]) — Broad skill library across ideation, literature, model work, evaluation, MLOps, agents, RAG, and paper writing for Claude Code, Codex, and Gemini-style hosts.
• academic-research-skills (2026, [CLI-native]) — Claude Code skill suite for academic research, writing, review, revision, finalization, response-to-reviewer artifacts, integrity reports, reviewer simulation, and reproducibility checks.
• claude-code-my-workflow (2026, [CLI-native]) — Personal Claude Code setup for papers, slides, data analysis, replication packages, preregistration, review, and referee responses; notable "contractor mode" and adversarial QA loops.
• claude-scholar (2026, [CLI-native]) — Semi-automated research assistant for CS/AI across literature, coding, experiments, reports, writing, rebuttal, and knowledge management; skills + commands + agents + rules + hooks + Zotero + optional Obsidian.
• W&B Skills (2026, [CLI-native]) — Installable Claude Code / Codex skills for W&B experiments, traces, evals, and reports (training, run comparison, failure-mode analysis, report generation).
• Organon (2026, [CLI-native]) — Claude Code scientific OS covering literature search, hypothesis testing, data analysis, manuscript drafting, scheduled jobs, and science communication, with verification gates that block fabricated citations and unsupported claims.
• MedSci Skills (2026, [CLI-native]) — Medical-science Claude Code skill suite (~39 skills) across topic discovery, literature search, study design, statistics, figures, writing, compliance, peer review, revision, and presentations with PubMed/Semantic Scholar/CrossRef-verified citations.
• Claude Code for an Academic Researcher (2026, [CLI-native]) — Full Claude Code infrastructure for academics with 48 skills, 6 agents, 9 hooks, rules, context libraries, LaTeX, bibliography tooling, Notion integration, and reviewer roles (domain reviewer, paper critic, peer reviewer, referee).
• scientific-agent-skills (2026, [CLI-compatible]) — Agent Skills catalog across biology, chemistry, medicine, ML, materials, physics, engineering, data analysis, and scientific communication with Cursor / Claude Code / Codex compatibility.
• ai-skill-scholar (2026, [CLI-compatible]) — Three skills for scholarly search, citation analysis, and literature review via OpenAlex; two-pass orchestrator with persistent session state.
• EvoSkills (2026, [CLI-compatible]) — Installable research-skill library for ideation, paper planning, experiment pipelines, writing, review, rebuttal, memory, paper navigation, and surveys across Claude Code, Codex, Gemini CLI, Cursor, OpenCode, and DeepAgents.
• ML Experiment Tracking Setup skill (2026, [CLI-compatible]) — Skill for scaffolding MLflow or W&B experiment-tracking before agent-driven iteration; verify adoption before depending on it.
• ResearchClaw (2025, [CLI-compatible]) — Personal research assistant exposing extensible skills for retrieval, analytics, code execution, and document understanding.

Literature Discovery and Evidence Grounding

• paper-search-mcp (2026, [CLI-native]) — Searches and downloads papers from arXiv, PubMed, bioRxiv via MCP tools and a Claude Code skill; discovery layer only — ranking discipline still needed.
• LitLLM (2025, [CLI-native]) — Literature-review toolkit and Claude Code skill for retrieving, ranking, and synthesizing related work over OpenAlex, Google Scholar, embedding search, and LLM reranking.
• PaperQA2 (2024, [CLI-compatible]) — Scientific RAG over papers, text, Office files, and code with cited evidence; callable as a literature-QA backend for related work, contradiction checks, paper writing, and rebuttals. [](https://github.com/future-house/paper-qa)
• GPT Researcher (2024, [CLI-compatible]) — Long-form cited research reports over web and local sources with planning, crawling, source tracking, MCP, and Claude Skill install; scholarly claims still need verification.
• Zotero MCP / ZotPilot (2025, [CLI-compatible]) — Zotero library, citations, summaries, PDFs, and annotations via MCP; ZotPilot adds semantic search and lit-review drafting over local data.
• semantic-scholar-mcp (2026, [CLI-compatible]) — Lightweight MCP server for searching and analyzing papers through the Semantic Scholar API; unofficial and small.
• arxiv-mcp-server (2025, [CLI-compatible]) — Search, download, store, and analyze arXiv papers via MCP, with security guidance on adversarial paper text and prompt injection.
• PubMed MCP Server (rolling, [CLI-compatible]) — PubMed / NCBI E-utilities MCP with article search, retrieval, research-plan generation, filtering, citation analysis, metadata visualization, and biomedical research-agent scaffolding.
• ChatPaper (2023, [CLI-compatible]) — Open-source paper-summarization and literature-synthesis tool widely adopted as an early LLM-assisted reading workflow.
• Tongyi DeepResearch (2025, [CLI-compatible]) — Alibaba open-source agentic LLM specialized for long-horizon deep information seeking and report synthesis.
• OpenScholar (2024, [adjacent]) — Retrieval-augmented scholarly assistant over a large open-access corpus with ScholarQABench-style evaluation; canonical grounding layer, not a CLI workflow. [](https://github.com/akariasai/openscholar)
• STORM (2024, [adjacent]) — Long-form grounded articles via perspective discovery, simulated writer-expert chat, outline organization, and drafting from retrieved sources; Co-STORM adds collaborative discourse. [](https://github.com/stanford-oval/storm)
• Gemini Deep Research Agent (2026, [adjacent]) — API-level deep-research agent for planning, executing, and synthesizing multi-step research with background execution, planning viz, MCP, and document input.
• OpenNovelty (2026, [adjacent]) — Agentic novelty-assessment pipeline that extracts claims, retrieves prior work, builds a hierarchical taxonomy, compares contributions against full texts, and produces evidence-grounded novelty reports. [](https://github.com/january-blue/OpenNovelty)
• LitLLM (paper) (2024, [adjacent]) — Literature-review toolkit paper integrating LLMs with academic databases for dense retrieval over scientific corpora.
• PaSa (2025, [adjacent]) — LLM agent for academic search that issues follow-up queries and iteratively refines candidate paper sets like a human probing a topic.
• AutoSurvey (2024, [adjacent]) — Single-pass automated survey-writing system establishing baseline feasibility for turning retrieved literature into structured drafts.
• SurveyX (2025, [adjacent]) — Academic survey-automation system evaluated separately on content quality, structure quality, and citation accuracy.
• SurveyForge (2025, [adjacent]) — Survey-writing system learning outline heuristics from human surveys and using memory-driven content generation.
• LiRA (2025, [adjacent]) — Multi-agent literature-review generator splitting retrieval, verification, organization, and writing into specialized roles.
• Agentic AutoSurvey (2025, [adjacent]) — Agentic survey-generation workflow with dedicated agents for retrieval, organization, and narrative synthesis.
• IterSurvey (2025, [adjacent]) — Literature-survey agent treating outline generation as iterative planning with stability checks.
• InteractiveSurvey (2025, [adjacent]) — Interact

…

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: chrisliu298
• Source: chrisliu298/awesome-research-agents
• License: CC0-1.0

Install and usage instructions live in the source repository linked above.