# GPT Researcher — Open-source autonomous research agent that writes cited long-form reports from web and local sources. > Source: The Agents Index — https://theagentsindex.com/gpt-researcher (structured, researched, re-verified) > Facts last verified: 2026-09-16 GPT Researcher is an open-source autonomous research agent that plans a topic into subqueries, validates web and local sources in parallel, and writes long-form reports with inline citations. It ships as a Python package, a self-hosted FastAPI server, a Docker image and an MCP server. | Fact | Value | | --- | --- | | Website | https://gptr.dev | | Price | GPT Researcher is published under the MIT License. There is no SaaS subscription, no API gateway, and no rate limit on this site. | | Free tier | Yes | The whole product is the free tier: MIT-licensed package and MCP server, self-hosted, with the vendor stating "None imposed by GPT Researcher. Limits come from your underlying provider rate limits." You supply and pay for LLM and retriever keys, and need Python 3.11+ plus somewhere to run it. ## What it does **Done-for-you service** · No — You run the software yourself; no team builds or operates it for you, and managed VPC deployment or an SLA exists only as an invitation to email or Discord. > No first-party hosted SaaS today. > — https://gptr.dev/pricing · Enterprise / managed section ## How much it does unattended **Runs autonomously** · Yes — A run plans subtopics, searches, validates sources and writes the report with no human step, and the human-feedback gate is opt-in. > include_human_feedback - If true, the user can provide feedback to the agents. If false, the agents will work autonomously. > — https://docs.gptr.dev/docs/gpt-researcher/multi_agents/langgraph · Task.json fields list, include_human_feedback **Multi-agent** · Yes — A LangGraph flow coordinates a chief editor, researcher, reviewer, revisor, writer and publisher, and an AG2 integration is documented too. > By using Langgraph, the research process can be significantly improved in depth and quality by leveraging multiple agents with specialized skills. > — https://docs.gptr.dev/docs/gpt-researcher/multi_agents/langgraph · Use case section **Agent permissions** · Partly — There are no approval prompts mid-run; you constrain it beforehand through configuration such as domain allow-lists, report source and depth, and its actions stay read-only research. > You can filter web search results by specific domains when using either the Tavily or Google Search retrievers. > — https://docs.gptr.dev/docs/gpt-researcher/context/filtering-by-domain · Filtering by Domain, opening paragraph ## Which models it runs on **Claude** · Yes — Claude is named as a supported model and anthropic appears in the docs list of LLM providers, with your own Anthropic key. > Thanks to the massive community contributions, GPT Researcher supports over 100 different LLMs such as Gemini, Claude, Llama, etc. > — https://gptr.dev/ · FAQ, 'What LLMs does GPT Researcher support?' **GPT** · Yes — OpenAI is the default provider for the fast, smart and strategic model slots, so GPT models work out of the box. > GPT Researcher defaults to our recommended suite of integrations: OpenAI for LLM calls and Tavily API for retrieving real-time web information. > — https://docs.gptr.dev/docs/gpt-researcher/gptr/config · Configuration, introductory paragraph **Open models** · Yes — Local open-weight models run through Ollama, and the supported-provider list also includes huggingface, vllm, together and deepseek. > If you deploy ollama locally, a .env like so, should enable powering GPT-Researcher with Ollama: > — https://docs.gptr.dev/docs/gpt-researcher/llms/running-with-ollama · Querying your Custom LLM with GPT-Researcher section **Model choice** · Yes — You choose the provider and model per role through FAST_LLM, SMART_LLM and STRATEGIC_LLM config keys rather than accepting a vendor default. > Easily connect any LLM of your choice from OpenAI, Anthropic, Groq to Llama 3 and HF. > — https://gptr.dev/ · 'Tailored just for you' features grid, Support for all LLMs card ## Where you use it **In your editor** · Yes — The MCP server registers over stdio in Cursor and Claude Desktop, and a Claude Skill extends the agent into Claude Code. > It is self-hosted (MIT, Python 3.11+) and supports stdio (Claude Desktop, Cursor), SSE (Docker / n8n / web), and Streamable HTTP. > — https://gptr.dev/agent · MCP server section **On the command line** · Yes — A documented cli.py takes the query, report type and tone as arguments and writes the finished report to an outputs directory. > This command-line interface (CLI) tool allows you to generate research reports using the GPTResearcher class. > — https://docs.gptr.dev/docs/gpt-researcher/getting-started/cli · Run with CLI, opening paragraph **In a browser** · Yes — The project ships its own browser frontends (static, NextJS, embed script) that you host, and a vendor-run instance answers at app.gptr.dev. > Includes an easy-to-use web interface (HTML/CSS/JS) > — https://docs.gptr.dev/docs/gpt-researcher/getting-started/introduction · Features list ## Whose machine it runs on **Self-hosted** · Yes — Self-hosting is the only supported deployment: pip package, FastAPI server, docker-compose or Linux host, all on infrastructure you own. > You self-host the agent and pay only your underlying LLM and search-API providers. > — https://gptr.dev/pricing · Free and open source, intro paragraph **Open source** · Yes — The core agent and the MCP server are MIT-licensed public repositories, published to PyPI and npm with community contributions. > GPT Researcher is published under the MIT License. > — https://gptr.dev/pricing · Free and open source, intro paragraph ## What it costs to run **How it meters** · Partly — The vendor bills nothing at all; cost is per research run of LLM and retriever spend, and its own published estimates range widely. > Each deep research takes around 5 minutes to complete and costs around $0.4 (using o3-mini on "high" reasoning effort) > — https://docs.gptr.dev/docs/gpt-researcher/gptr/deep_research · Deep Research, opening section **Free tier** · Yes — No call, no card and no account: clone or pip install the MIT package and supply your own provider keys to start. > There is no SaaS subscription, no API gateway, and no rate limit on this site. > — https://gptr.dev/pricing · Free and open source, intro paragraph **API access** · Yes — Your own stack drives it through the self-hosted FastAPI REST endpoints, an OpenAPI 3.1 spec, SSE streaming and signed outbound webhooks. > The self-hosted GPT Researcher FastAPI server exposes a small, agent-friendly REST API. > — https://gptr.dev/api-reference · API reference, page header paragraph **MCP server** · Yes — The official gptr-mcp server exposes deep_research, quick_search, write_report and source tools over stdio, SSE and Streamable HTTP. > The GPT Researcher MCP Server provides Model Context Protocol (MCP) integration for GPT Researcher, allowing AI assistants to perform autonomous, comprehensive web research and generate reports via the MCP protocol. > — https://docs.gptr.dev/docs/gpt-researcher/mcp-server/getting-started · MCP Server, Getting Started opening paragraph **Bring your own key** · Yes — Bring-your-own-key is the only model: your LLM and retriever credentials go in environment variables on infrastructure you run. > Provide your own LLM (e.g. OPENAI_API_KEY) and retriever (e.g. TAVILY_API_KEY) credentials via environment variables when you run the server or MCP server. > — https://gptr.dev/agent · Authentication section ## Buying it for a team **A company can buy it** · Partly — An organisation can adopt the MIT software freely but cannot buy anything: there is no published contract, seat or subscription path, only an enterprise enquiry. > Both! GPT Researcher is ideal for anyone who needs to conduct research to make unbiased and informed decisions. We work closely with individuals, startups and enterprises > — https://gptr.dev/ · FAQ, 'Is GPT Researcher suitable for individual or enterprise needs?' **Seat model** · No — One MIT-licensed tier priced at $0 for the whole product; no seats, user counts or subscription anywhere on the pricing page. > There is no SaaS subscription, no API gateway, and no rate limit on this site. > — https://gptr.dev/pricing · Pricing page, 'Free and open source' paragraph above the single 'Open Source - $0' plan card **Pooled budget** · No — There is no vendor-side balance; spend sits entirely with the LLM and retriever accounts you own. > None imposed by GPT Researcher. Limits come from your underlying LLM / retriever providers' rate limits. > — https://gptr.dev/pricing.md · Frequently asked questions, Are there usage limits? **Admin controls** · Partly — Whoever runs the deployment constrains the agent through config and environment variables, but there is no admin console, role separation or policy enforced above the operator. > MAX_SEARCH_RESULTS_PER_QUERY: Maximum number of search results to retrieve per query. Defaults to 5. > — https://docs.gptr.dev/docs/gpt-researcher/gptr/config · Configuration, supported options list **Audit log** · No — Each run writes a downloadable JSON event log of research steps, but no administrative trail of users, permissions or policy changes exists. > There is no GPT Researcher account system. > — https://gptr.dev/llms-full.txt · Section 6, Authentication and access **Single sign-on** · No — No SAML or OIDC is offered and there is no vendor login to federate; identity is whatever your own hosting puts in front of the server. > GPT Researcher is self-hosted; there is no gptr.dev account. > — https://gptr.dev/agent · Authentication section ## What happens to your code **Opt out of training** · Yes — Nothing reaches a GPT Researcher service to train on; only the providers you pick see your queries, and a local model plus SearXNG removes even that. > Only via the LLM and retriever providers you configure. Self-host with a local model and SearXNG to keep everything on-prem. > — https://gptr.dev/pricing · FAQ, 'Will my data leave my machine?' **Data residency** · Yes — Residency is wherever you deploy, since the vendor stores nothing; the jurisdictions that matter are your host's and your chosen providers'. > Self-host on your own machine, server, or container. > — https://gptr.dev/pricing · Open Source - $0 plan bullets **Getting out** · Yes — Reports export to PDF, Word, Markdown, JSON and CSV, run logs download as JSON, and no credits exist that could expire. > Export your research to any format from PDF, Word, Markdown to JSON and CSV. > — https://gptr.dev/ · 'Tailored just for you' features grid, Export to any Format card _Not established: Certifications_ ## Platform record (as of 2026-08-30) Reported by the code-hosting platform, not by the vendor. | Fact | Value | | --- | --- | | Licence | Apache-2.0 | | Stars | 29205 | | Forks | 3964 | | Last push | 2026-08-27 | ## Provenance 1 of 18 sources behind the facts above are independent of the vendor. ## Coverage elsewhere Pages about this product that the vendor does not control. The label states how independent each one is. | Source | Kind | Read | | --- | --- | --- | | arxiv.org | Independent source | 2026-08-23 | | nebius.com | Independent source | 2026-07-21 | ## What changed - 2026-09-16 — Capabilities · Commercial terms · Free tier · Price - 2026-09-02 — Capabilities · Commercial terms · Free tier - 2026-08-31 — Capabilities · Commercial terms · Free tier - 2026-08-22 — Alternatives · Documentation link · One-liner · Summary - 2026-08-19 — Summary ## Correcting this record Anything marked "Not established" is a question we have not answered — it is not a No. If you represent GPT Researcher, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth. Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.