Skip to content
Back to search

DeepResearch_AI

Open-source multi-agent research assistant: cited web reports in ~2 min, or a full academic paper from a 25-agent pipeline with human checkpoints. Contributors welcome.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/Aryan-Pardeshi/CiteProof-AI-MCP with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/Aryan-Pardeshi/CiteProof-AI-MCP'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/aryan-pardeshi--deepresearch-ai/badge.svg)](/listings/aryan-pardeshi--deepresearch-ai)

HTML

<a href="/listings/aryan-pardeshi--deepresearch-ai"><img src="/api/listings/aryan-pardeshi--deepresearch-ai/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

1. Clone & Configure

git clone https://github.com/Aryan-Pardeshi/DeepResearch_AI.git
cd DeepResearch_AI
cp .env.example .env

Minimum `.env` configuration:

LLM_API_KEY="your_api_key_here"
LLM_BASE_URL="https://api.deepseek.com"
TAVILY_API_KEY="your_tavily_key_here"
OPENALEX_EMAIL="you@example.com"
NCBI_EMAIL="you@example.com"
NCBI_TOOL_NAME="your_tool_name"

Optional academic API configuration:

SEMANTIC_SCHOLAR_API_KEY="your_api_key_here"
NCBI_API_KEY="your_api_key_here"
CORE_API_KEY="your_api_key_here"
ORCID_AUTHOR_ENRICHMENT=0

2. Run with Docker

docker compose up -d --build

Access the UI at **[http://localhost:8000](http://localhost:8000)**.

3. Local Development

python -m venv .venv
.venv\Scripts\activate      # Windows (or source .venv/bin/activate on Unix)
pip install -r requirements.txt
uvicorn backend.app.main:app --reload --port 8000

---

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 413 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # <img src="frontend/assets/logo.png" alt="" width="32" height="32" style="vertical-align: middle; border-radius: 7px;"> CiteProof
  2. **Anti-hallucination checks for AI literature reviews.** CiteProof checks recorded quotes and numbers against the source text, looks up DOIs, matches citations to a reference list, and checks that counts and risk-of-bias judgements hold up. It comes in two parts: an **MCP server** for Claude Science and other research agents, and a full **research assistant app** that writes evidence-grounded papers.
  3. [![npm](https://img.shields.io/npm/v/citeproof-mcp?label=npm%20citeproof-mcp)](https://www.npmjs.com/package/citeproof-mcp)
  4. [![Release](https://img.shields.io/github/v/release/Aryan-Pardeshi/CiteProof-AI-MCP)](https://github.com/Aryan-Pardeshi/CiteProof-AI-MCP/releases/latest)
  5. [![CI](https://github.com/Aryan-Pardeshi/CiteProof-AI-MCP/actions/workflows/mcp-npm.yml/badge.svg)](https://github.com/Aryan-Pardeshi/CiteProof-AI-MCP/actions/workflows/mcp-npm.yml)
  6. [![License: MIT](https://img.shields.io/badge/license-MIT-blue)](LICENSE)
  7. [![olud.ai](https://olud.ai/badge.php?tool=aryan-pardeshi-citeproof-ai-mcp)](https://olud.ai/project/aryan-pardeshi-citeproof-ai-mcp.html)
  8. **Research Web → [CiteProof.top](https://citeproof.top/)**
  9. **Claude Science MCP** → Customize → Connectors → Add connector → Local command:
  10. ```bash
  11. npx -y citeproof-mcp
  12. ```
  13. <p align="center">
  14. <a href="https://citeproof.top/"><img src="docs/media/citeproof-ad.gif" alt="CiteProof: PRISMA screening, DOI and retraction checks, claim checks against source text, and the CiteProof MCP inside Claude Science" width="800"></a>
  15. </p>
  16. ## Contents
  17. - [What is CiteProof?](#what-is-citeproof)
  18. - [CiteProof MCP](#citeproof-mcp)
  19. - [What it checks](#what-it-checks)
  20. - [Quick start](#quick-start)
  21. - [Companion skill](#companion-skill)
  22. - [The 40 tools](#the-40-tools)
  23. - [What it does not do](#what-it-does-not-do)
  24. - [CiteProof Research (the app)](#citeproof-research-the-app)
  25. - [Two modes](#two-modes)
  26. - [Architecture](#architecture)
  27. - [Research pipeline](#research-pipeline)
  28. - [3 human-in-the-loop quality gates](#3-human-in-the-loop-quality-gates)
  29. - [Academic integrity and research invariants](#academic-integrity-and-research-invariants)
  30. - [Tech stack](#tech-stack)
  31. - [Academic API coverage](#academic-api-coverage)
  32. - [Web search providers](#web-search-providers)
  33. - [Getting started](#getting-started)
  34. - [Repository layout](#repository-layout)
  35. - [Contributing](#contributing)
  36. - [Licence](#licence)
  37. - [Star history](#star-history)
  38. ---
  39. ## What is CiteProof?
  40. AI research agents are now good at finding and reading papers. They are less good at proving their work: which databases were searched, how many records were screened, whether a quoted sentence is really in the paper, whether two reviewers agreed. CiteProof is built around that gap.
  41. | | **CiteProof MCP** | **CiteProof Research** |
  42. |---|---|---|
  43. | **What it is** | An MCP server with 40 evidence and audit tools | A web app with a 24-agent research pipeline |
  44. | **Who does the research** | Your agent (Claude Science, or any MCP client) | The app's own agents |
  45. | **Language model inside** | None. It needs no API key | Yes, any OpenAI-compatible endpoint |
  46. | **Best for** | Systematic reviews that need an audit trail | Fast cited reports and full evidence-grounded papers |
  47. | **Where it lives** | [`MCP/`](MCP/) | `backend/`, `frontend/` |
  48. A small first benchmark (two reviews against Claude Science) showed that agents with strong search find far more of the known included studies than CiteProof Research does. That is why the MCP server does not try to search: it audits the work of the agent that does.
  49. ---
  50. ## CiteProof MCP
  51. CiteProof MCP adds evidence checks to a research agent. Language models can invent quotes, figures and references, so each quote is compared with the source text you supply, each number with its quote, each DOI with Crossref (with a doi.org fallback), and each author-year or DOI citation with the reference list you supply (numeric citations such as `[1]` are counted, not resolved). Around those checks it keeps a protocol, PRISMA counts, a screening log, an evidence table, claim checks, risk-of-bias records and an exportable audit trail. It flags what does not hold and never decides whether a study is right. It works with [Claude Science](MCP/docs/CLAUDE_SCIENCE.md) and with other tools that can run a stdio MCP server, such as research agents and desktop AI clients.
  52. The server has no language model inside. The agent searches, reads, decides and writes. The server stores each decision, checks the evidence it points to, computes the numbers and refuses bad input.
  53. Full documentation: [MCP/README.md](MCP/README.md) · tool reference: [MCP/docs/SPEC.md](MCP/docs/SPEC.md).
  54. ### What it checks
  55. | Check | How |
  56. |---|---|
  57. | A quote really appears in the source text | The agent sends the quote and the text. The server matches them after normalising case, whitespace, quotes, dashes and line-break hyphens, on word boundaries. The result is stored as `quote_verified`. |
  58. | Numbers in the evidence table come from the quote | Sample size, estimate and confidence limits must be written in the quote. Estimates outside their interval are refused. |
  59. | PRISMA numbers add up | Counts come from the recorded searches, the deduplication step and the screening decisions, and the flow invariants are checked on every call. |
  60. | Two reviewers agree | Cohen's kappa over the papers both reviewers decided, with the disagreements listed. |
  61. | DOIs exist and are not retracted | Crossref and doi.org lookups. An unreachable registry gives "unverified", never "not found". |
  62. | Citations match the reference list | In-text citations are matched by DOI, author and year, or title. "Resolved" means matched, not supported. |
  63. | Risk-of-bias overall judgements follow the published rules | RoB 2, ROBINS-I and Newcastle-Ottawa (AHRQ thresholds). |
  64. | Duplicates | Deterministic clustering by DOI, PMID, then title plus year plus first author. |
  65. ### Quick start
  66. One command, in any MCP client that runs local (stdio) servers:
  67. ```bash
  68. npx -y citeproof-mcp
  69. ```
  70. The [`citeproof-mcp`](https://www.npmjs.com/package/citeproof-mcp) npm package is a small launcher. It finds [uv](https://docs.astral.sh/uv/) (or installs it, unless `CITEPROOF_NO_AUTO_INSTALL=1`), then runs the Python server from the wheel attached to the matching [GitHub release](https://github.com/Aryan-Pardeshi/CiteProof-AI-MCP/releases/latest). You need Node.js 18 or newer. Python and git are not required: uv downloads Python 3.12 when needed.
  71. | Client | How to add CiteProof MCP | Status |
  72. |---|---|---|
  73. | **Claude Science** | Customize → Connectors → Add connector → **Local command**. Name `citeproof`, command `npx -y citeproof-mcp`. | npx install tested on Claude Science 0.1.70 (Windows 11, 24 tools listed for 0.2.1; 0.3.0 with 40 tools not yet tested in Claude Science); bundle install tested on 0.1.67 |
  74. | **Claude Desktop** | Download [`citeproof.mcpb`](https://github.com/Aryan-Pardeshi/CiteProof-AI-MCP/releases/latest/download/citeproof.mcpb) and open it (Settings → Extensions), or add the JSON below to `claude_desktop_config.json`. | New in 0.2.0 |
  75. | **Claude Code** | `claude mcp add citeproof -- npx -y citeproof-mcp` | New in 0.2.0 |
  76. | **Cursor, VS Code, Windsurf, other clients** | Add the JSON below to the client's MCP settings (VS Code: the `servers` form further down). | Not yet tested |
  77. ```json
  78. {
  79. "mcpServers": {
  80. "citeproof": {
  81. "command": "npx",
  82. "args": ["-y", "citeproof-mcp"]
  83. }
  84. }
  85. }
  86. ```
  87. VS Code's built-in MCP support (`.vscode/mcp.json` or your user MCP settings) uses `servers` instead of `mcpServers`:
  88. ```json
  89. {
  90. "servers": {
  91. "citeproof": {
  92. "type": "stdio",
  93. "command": "npx",
  94. "args": ["-y", "citeproof-mcp"]
  95. }
  96. }
  97. }
  98. ```
  99. Everything runs on your computer. Review files go to `CITEPROOF_MCP_HOME`, or to `~/.citeproof-mcp` when it is not set. The server's only network calls are DOI and retraction lookups to Crossref and doi.org. On first run the launcher also downloads its own files: the npm package, and, when no compatible Python is found, uv, Python 3.12 and the server wheel. claude.ai custom connectors need a remote server, so they are not supported.
  100. If Claude Science reports that a file is not readable inside its sandbox, set `CITEPROOF_UV_HOME` (Advanced settings → environment variables) to a folder inside the connector's workspace, so uv keeps its cache and Python there. The older bundle install in [MCP/docs/CLAUDE_SCIENCE.md](MCP/docs/CLAUDE_SCIENCE.md) still works.
  101. <details>
  102. <summary>Install from source with pip</summary>
  103. ```bash
  104. pip install ./MCP
  105. ```
  106. Then point the client at the `citeproof-mcp` command (stdio), optionally with `"env": { "CITEPROOF_MCP_HOME": "/path/to/your/review-data" }`.
  107. </details>
  108. ### Companion skill
  109. [`MCP/skills/citeproof-review`](MCP/skills/citeproof-review/SKILL.md) teaches an agent the review workflow and the tool order. Install steps for Claude Code, Claude Desktop and Claude Science are in [MCP/skills/README.md](MCP/skills/README.md). A reviewer-instructions snippet for Claude Science is in [MCP/docs/REVIEWER_SNIPPET.md](MCP/docs/REVIEWER_SNIPPET.md).
  110. ### The 40 tools
  111. | Group | Tools |
  112. |---|---|
  113. | Set up and status | `review_create`, `review_list`, `protocol_record`, `review_status` |
  114. | Searches and duplicates | `prisma_record_search`, `dedup_papers`, `prisma_record_dedup` |
  115. | Search validation | `recall_seed_record`, `recall_check` |
  116. | Screening | `screening_record`, `screening_record_many`, `screening_list`, `screening_agreement`, `reviewer_record`, `screening_stage_set`, `screening_stage_close` |
  117. | Paper and study identity | `paper_resolve`, `paper_alias_record`, `study_link`, `study_unlink`, `study_overview` |
  118. | PRISMA | `prisma_status`, `prisma_flow_diagram`, `prisma_checklist`, `prisma_excluded_studies` |
  119. | Evidence and bias | `evidence_record`, `evidence_table`, `claim_record`, `claim_report`, `rob_record`, `rob_summary` |
  120. | Statistics | `stats_check_ratio_ci`, `stats_recompute_2x2`, `stats_check_evidence` |
  121. | Citations, manuscript and prose | `verify_dois`, `resolve_citations`, `lint_prose`, `manuscript_check` |
  122. | History and export | `review_history`, `export_bundle` |
  123. ### What it does not do
  124. - It does not search for papers. Use your agent's search, or the CiteProof Research app below.
  125. - It does not judge anything. A label such as "supported" is the agent's judgement. The server only checks that the quote is in the text.
  126. - It does not read PDFs. The agent supplies the text.
  127. - The PRISMA checklist and the report number check are keyword checks. They show what is mentioned, not whether it is reported well.
  128. - Blinding and reviewer types are by convention, with an audit trail, not a guarantee. The server cannot verify caller identity or authenticate models; AI screening is never called independent or verified. Two runs of the same model are not independent reviewers.
  129. - Statistical and manuscript checks report detected inconsistencies, never that a paper is wrong; seed recall is not proof that a search was exhaustive. The server contains no language model.
  130. ---
  131. ## CiteProof Research (the app)
  132. An AI-powered multi-agent research workspace with **two modes**: fast web research that returns a cited report, and a full evidence-grounded academic pipeline that writes complete research papers whose cited claims link back to source passages, with human-in-the-loop quality gates.
  133. ### Two modes
  134. | | **DeepSearch** | **Research Mode (Evidence-First)** |
  135. |---|---|---|
  136. | **Goal** | Answer a question from the live web | Write a full, evidence-grounded academic paper |
  137. | **Sources** | Tavily web search | OpenAlex · Semantic Scholar · Crossref · PubMed · arXiv · Europe PMC · DOAJ · DataCite · OpenCitations |
  138. | **Agents** | 5 (fan-out parallel researchers) | 24 specialized agents across 5 evidence phases |
  139. | **HITL Checkpoints** | 1 (plan approval) | 3 Quality Gates (Protocol, Evidence Corpus, Hypotheses) |
  140. | **Data Invariant** | LLM web summary | Deterministic `EvidenceRecord` store (quotes, metrics, baselines) |
  141. | **PRISMA Tracking** | N/A | Deterministic PRISMA 2020 flow counts & assertions |
  142. | **Integrity Audit** | Citation regex | Automated citation & numerical claim grounding verification |
  143. | **Runtime** | ~1-2 min | ~5-15 min |
  144. | **Output** | Structured markdown report | Full paper + PRISMA flow chart + evidence mapping matrix |
  145. | **Export** | `.md` | `.pdf` · `.docx` |
  146. ### Architecture
  147. Both modes run as separate LangGraph state machines behind one FastAPI backend, sharing the LLM layer, checkpointer, and SSE streaming transport.
  148. ```mermaid
  149. flowchart TB
  150. UI["🖥️ Frontend<br/>Vanilla JS · Swiss Modernism 2.0"]
  151. API["⚡ FastAPI<br/>SSE streaming & Heartbeats"]
  152. UI <-->|"Server-Sent<br/>Events"| API
  153. API --> DS["🔎 DeepSearch<br/>Graph"]
  154. API --> RM["📚 Research Mode<br/>24-Agent Graph"]
  155. DS --> TAV["🌐 Tavily<br/>Web Search"]
  156. RM --> ACAD["🎓 Multi-Source Academic APIs<br/>8 discovery providers · citation expansion<br/>author enrichment · full-text resolution"]
  157. DS --> LLM["🧠 LLM Layer<br/>Structured JSON Outputs"]
  158. RM --> LLM
  159. LLM --> CACHE[("💾 SQLite<br/>response cache")]
  160. DS --> CP[("🔁 Checkpointer<br/>resumable state")]
  161. RM --> CP
  162. style UI fill:#4f46e5,stroke:#3730a3,color:#fff
  163. style API fill:#4f46e5,stroke:#3730a3,color:#fff
  164. style DS fill:#0891b2,stroke:#0e7490,color:#fff
  165. style RM fill:#0891b2,stroke:#0e7490,color:#fff
  166. style TAV fill:#ea580c,stroke:#c2410c,color:#fff
  167. style ACAD fill:#ea580c,stroke:#c2410c,color:#fff
  168. style LLM fill:#db2777,stroke:#be185d,color:#fff
  169. style CACHE fill:#475569,stroke:#334155,color:#fff
  170. style CP fill:#475569,stroke:#334155,color:#fff
  171. ```
  172. ### Research pipeline
  173. **Research question → Scope & protocol → Literature retrieval → Evidence extraction → Synthesis → Validation & export**
  174. `24 specialized agents · 5 phases · 3 human approval gates`
  175. ### 3 human-in-the-loop quality gates
  176. Each gate genuinely pauses graph execution via LangGraph `interrupt()`, allowing the author to approve or request targeted revisions in natural language:
  177. | Gate | Phase | Review Artifacts | Guarantees |
  178. |---|---|---|---|
  179. | **Gate 1: Planning & Protocol** | Phase 1 | PICOC Protocol, Objectives, Research Questions, Boolean Keywords | Prevents query divergence before initiating multi-source retrieval |
  180. | **Gate 2: Evidence & Corpus** | Phase 3 | PRISMA Flow Tracker, Screened Literature, Extracted Evidence Records | Verifies empirical grounding before theoretical synthesis begins |
  181. | **Gate 3: Hypotheses & Framework** | Phase 4 | Conceptual Framework, Research Gaps, Formulated Hypotheses (H1..H5) | Confirms theoretical validity before committing to full paper generation |
  182. ### Academic integrity and research invariants
  183. 1. **Deterministic Flow-Tracking State Machine (PRISMA 2020-aligned)**:
  184. `PRISMATracker.validate_invariants()` asserts these *internal* conservation relations:
  185. ```python
  186. records_identified - duplicates_removed == records_after_dedup
  187. records_screened - excluded_title_abstract == full_text_requested
  188. full_text_assessed - excluded_full_text == studies_included
  189. ```
  190. `full_text_requested` is the tracker's internal shortcut for PRISMA's *reports sought for retrieval*. The distinct PRISMA stages *reports not retrieved* (`full_text_unavailable`) and *reports assessed for eligibility* (`full_text_assessed`) are tracked as separate fields, but the hand-off between them is not asserted—so the equations above are internal invariants rather than the complete PRISMA 2020 flow. Every count is tracked deterministically in code—never generated or estimated by an LLM.
  191. 2. **Immutable Evidence Store**:
  192. - `PaperRecord.paper_id`: first 16 hex characters of a SHA-256 digest (a truncated representation, not the full digest) computed over a source-qualified key: `doi:<canonicalized DOI>` when a DOI is available, otherwise `title:<normalized title>:<year>`. Normalized titles are not assumed unique, so the fallback key also folds in the publication year (`nd` when unknown).
  193. - `EvidenceRecord.evidence_id`: deterministic composite identifier `{paper_id}_ev001` (the composite is not itself hashed) anchoring exact quote, metric, baseline value, and effect direction.
  194. - `ReviewClaim.claim_id`: deterministic composite identifier `{section}_cl001` (section slug truncated to 8 characters) linking paper claims to supporting evidence IDs.
  195. 3. **Deterministic Validation Pipeline**:
  196. - `citation_validator`: Resolves every in-text citation against known `PaperRecord` metadata and every `ReviewClaim` through its claim → evidence → paper → DOI/URL chain; flags citations and claims that do not resolve. This is a resolution check only: it does not test that the cited passage supports the sentence, so a resolving claim is marked `resolved`, never `verified`.
  197. - `claim_validator`: Matches numbers in quantitative sentences against extracted `EvidenceRecord` benchmark metrics (numeric match only).
  198. - `integrity_auditor`: Validates PRISMA invariants and issues comprehensive `ValidationReport`; `passed_all_gates` means the resolution and numeric-match gates passed, not that sources support the prose.
  199. ### Tech stack
  200. | Layer | Technology |
  201. |---|---|
  202. | **Orchestration** | LangGraph (StateGraph, `interrupt()` HITL, `Send()` fan-out, SQLite persistence) |
  203. | **Structured LLM** | Pydantic v2 schemas + `ainvoke_structured_with_retry` |
  204. | **Academic Sources** | OpenAlex, Semantic Scholar, Crossref, PubMed, arXiv, Europe PMC, DOAJ, DataCite, OpenCitations |
  205. | **Author Enrichment** | Optional ORCID author-name canonicalization and iD lookup |
  206. | **Full-Text Ingestion** | Unpaywall → Europe PMC → CORE fallback chain |
  207. | **Backend API** | FastAPI + Server-Sent Events (SSE) with keep-alive heartbeats |
  208. | **Frontend UI/UX** | Vanilla JS + CSS (Swiss Modernism 2.0, zero-build, responsive) |
  209. | **Figures & Charts** | matplotlib (PRISMA 2020 Flowchart, Hypothesis Evidence Matrix) |
  210. | **Document Export** | FPDF2 (`.pdf` with hanging APA indents) · python-docx (`.docx`) |
  211. | **MCP server** | `mcp` SDK (1.27 and 2.x), SQLite, Pydantic v2, httpx |
  212. ### Academic API coverage
  213. Research Mode uses each academic API for a specific stage:
  214. | API | Role | Configuration |
  215. |---|---|---|
  216. | **OpenAlex** | Literature discovery and open-access metadata | `OPENALEX_EMAIL` |
  217. | **Semantic Scholar** | Literature discovery and citation metadata | Optional `SEMANTIC_SCHOLAR_API_KEY` |
  218. | **Crossref** | Literature discovery and DOI metadata resolution | Uses `OPENALEX_EMAIL` for polite-pool requests |
  219. | **PubMed (NCBI E-utilities)** | Biomedical literature discovery | `NCBI_EMAIL` and `NCBI_TOOL_NAME`; optional `NCBI_API_KEY` |
  220. | **arXiv** | Preprint discovery | No credentials |
  221. | **Europe PMC** | Biomedical discovery and full-text fallback | No credentials |
  222. | **Directory of Open Access Journals (DOAJ)** | Open-access journal discovery | No credentials |
  223. | **DataCite** | Research-output and dataset discovery | No credentials |
  224. | **OpenCitations** | Forward and backward citation-graph expansion | No credentials |
  225. | **ORCID** | Optional author identity enrichment | Set `ORCID_AUTHOR_ENRICHMENT=1`; optional `ORCID_MAX_AUTHORS` (default 2, max 10) |
  226. | **Unpaywall** | Open-access PDF resolution | Uses `OPENALEX_EMAIL` |
  227. | **CORE** | Optional full-text fallback | Optional `CORE_API_KEY` |
  228. ### Web search providers
  229. DeepSearch's `search_web` tool goes through a small provider layer instead of a
  230. single Tavily client. `WEB_SEARCH_PROVIDERS` is an ordered, comma-separated list
  231. of provider names and defaults to `tavily`.
  232. - A provider is tried only when it is listed **and** its API key is set.
  233. Providers without credentials are skipped silently, the same way Research Mode
  234. skips an academic source that has no key.
  235. - If a provider errors, times out or returns no results, the next configured
  236. provider in the list is tried. If every provider fails, `search_web` returns an
  237. error string (`Error using <Provider> search: ...`) instead of raising, so the
  238. research graph keeps running.
  239. - API clients are created lazily on first use, so importing the module never
  240. requires a key.
  241. | Provider | Configuration | Paid? |
  242. |---|---|---|
  243. | **Tavily** — default | `TAVILY_API_KEY` | Paid service with a free tier |
  244. | **SerpBase** — optional Google organic results | `SERPBASE_API_KEY` and `serpbase` in `WEB_SEARCH_PROVIDERS` | Paid third-party service; new accounts get 100 free credits |
  245. SerpBase has no date filter, so a search with a `time_range` skips it and moves
  246. on to the next provider. SerpBase is never used unless it is both listed and keyed:
  247. ```bash
  248. WEB_SEARCH_PROVIDERS="tavily,serpbase"
  249. SERPBASE_API_KEY="your_key_here"
  250. ```
  251. With only `TAVILY_API_KEY` set, DeepSearch behaves exactly as it did before.
  252. ### Getting started
  253. #### 1. Clone and configure
  254. ```bash
  255. git clone https://github.com/Aryan-Pardeshi/CiteProof-AI-MCP.git
  256. cd CiteProof-AI-MCP
  257. cp .env.example .env
  258. ```
  259. Minimum `.env` configuration:
  260. ```bash
  261. LLM_API_KEY="your_api_key_here"
  262. LLM_BASE_URL="https://api.deepseek.com"
  263. TAVILY_API_KEY="your_tavily_key_here"
  264. OPENALEX_EMAIL="you@example.com"
  265. NCBI_EMAIL="you@example.com"
  266. NCBI_TOOL_NAME="your_tool_name"
  267. ```
  268. Optional academic API configuration:
  269. ```bash
  270. SEMANTIC_SCHOLAR_API_KEY="your_api_key_here"
  271. NCBI_API_KEY="your_api_key_here"
  272. CORE_API_KEY="your_api_key_here"
  273. ORCID_AUTHOR_ENRICHMENT=0
  274. ORCID_MAX_AUTHORS=2
  275. ```
  276. #### 2. Run with Docker
  277. ```bash
  278. docker compose up -d --build
  279. ```
  280. Access the UI at **[http://localhost:8000](http://localhost:8000)**.
  281. #### 3. Local development
  282. ```bash
  283. python -m venv .venv
  284. .venv\Scripts\activate # Windows (or source .venv/bin/activate on Unix)
  285. pip install -r requirements.txt
  286. uvicorn backend.app.main:app --reload --port 8000
  287. ```
  288. ---
  289. ## Repository layout
  290. | Folder | What is in it |
  291. |---|---|
  292. | [`MCP/`](MCP/) | CiteProof MCP: the server package `citeproof_mcp`, its tests and docs, the npm launcher (`MCP/npm`), the Claude Desktop extension (`MCP/mcpb`), the companion skill (`MCP/skills`) and the Claude Science install script |
  293. | `backend/` | CiteProof Research: FastAPI app, LangGraph agents, academic API adapters |
  294. | `frontend/` | The web UI (vanilla JS, no build step) |
  295. | `tests/` | Offline tests for the app (`pytest`) |
  296. | `docs/` | Design notes for the app |
  297. | `scripts/` | Helper scripts for the app |
  298. ---
  299. ## Contributing
  300. Contributions are welcome, from typo fixes to new academic source adapters and new MCP checks. Start with [CONTRIBUTING.md](CONTRIBUTING.md) for setup and a project map, then pick an issue labelled [`good first issue`](https://github.com/Aryan-Pardeshi/CiteProof-AI-MCP/labels/good%20first%20issue) or [`help wanted`](https://github.com/Aryan-Pardeshi/CiteProof-AI-MCP/labels/help%20wanted). The test suites run offline with `pytest`: run `pytest --ignore=MCP` from the repository root for the app, and `python -m pytest` inside `MCP/` for the MCP server.
  301. ---
  302. ## Licence
  303. MIT, see [LICENSE](LICENSE). Two files inside the MCP server are adapted from Anthropic's Claude Science under the Apache License 2.0; see [MCP/NOTICE.md](MCP/NOTICE.md).
  304. ---
  305. ## Star History
  306. <a href="https://www.star-history.com/?repos=aryan-pardeshi%2Fciteproof-ai-mcp&type=date&legend=top-left">
  307. <picture>
  308. <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/chart?repos=aryan-pardeshi/citeproof-ai-mcp&type=date&theme=dark&legend=top-left" />
  309. <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/chart?repos=aryan-pardeshi/citeproof-ai-mcp&type=date&legend=top-left" />
  310. <img alt="Star History Chart" src="https://api.star-history.com/chart?repos=aryan-pardeshi/citeproof-ai-mcp&type=date&legend=top-left" />
  311. </picture>
  312. </a>

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.