Deep Research MCP Server: Automated Research Reporting

Overview

Deep Research MCP Server is an implementation of the Model Context Protocol (MCP) focused on automating investigative and research workflows. It exposes a set of data-collection and analysis tools that an LLM or agent can call to explore web pages, PDFs, search results and internal vector stores, then synthesize structured research reports in the style of Grok/OpenAI/Gemini/Perplexity outputs.

The server is useful when you need reproducible, auditable research outputs: it centralizes source fetching, citation extraction, evidence aggregation, and report formatting so models can request and receive intermediate artifacts as well as final summaries. This makes it easier to build agents that produce traceable conclusions for literature reviews, due diligence, competitive analysis, or long-form summarization.

Features

• MCP-compatible tool registry for LLMs and agents
• Multi-source ingestion: web search (SerpAPI/Google), raw web pages, PDFs, and Git repositories
• Document parsing: OCR, HTML-to-text, structured metadata and citation extraction
• Vectorization and semantic search integration with common vector stores (e.g., Milvus, Weaviate, Pinecone)
• Report templates that produce structured Grok/OpenAI/Gemini/Perplexity-style outputs (Summary, Key Findings, Evidence, Sources, Next Steps)
• Task lifecycle management: asynchronous research tasks, progress events, and result retrieval
• Docker and local dev support, environment-driven configuration for API keys and tool toggles
• Audit-friendly output with source URLs, snippet offsets, and extraction timestamps

Installation / Configuration

Clone and run locally or via Docker. Replace environment variables with your API keys and desired services.

Clone and install

git clone https://github.com/reading-plus-ai/mcp-server-deep-research.git
cd mcp-server-deep-research
# Node.js example
npm install
cp .env.example .env
# edit .env to add API keys
npm run start

Docker

# build and run
docker build -t mcp-deep-research .
docker run -p 8080:8080 --env-file .env mcp-deep-research

Minimum .env (example)

PORT=8080
MCP_AUTH_TOKEN=your-secure-token
OPENAI_API_KEY=sk-...
SERPAPI_KEY=serp-...
VECTOR_DB_URL=https://your-vector-db.example
ENABLE_WEB_SCRAPER=true
DEFAULT_REPORT_STYLE=perplexity

Configuration notes

• Toggle ingestion tools (PDF, web_scraper, vector_store) using boolean env flags.
• Provide provider-specific keys (SERPAPI, BING, OPENAI, GEMINI) as needed.
• Configure vector DB URL and credentials for semantic search.

Available Tools / Resources

The server exposes these primary tools for MCP clients:

• web_search: query search engines and return ranked results (title, snippet, url)
• web_fetch: fetch and normalize HTML pages and extract text blocks
• pdf_parse: extract text and metadata from uploaded or remote PDFs
• semantic_search: nearest-neighbor queries against a configured vector store
• citation_extractor: find inline citations, DOIs, and structured references
• screenshot / render: capture rendered page snapshots for visual evidence
• repo_fetch: clone and extract README/code snippets from Git repos

Endpoint overview

API: Starting a Research Task (example)

Request to initiate a task:

curl -X POST "http://localhost:8080/v1/research" \
 -H "Authorization: Bearer $MCP_AUTH_TOKEN" \
 -H "Content-Type: application/json" \
 -d '{
   "query": "Recent advances in transformer sparsification for 2024",
   "scope": {"domains": ["arxiv.org","aclweb.org"], "maxPages": 50},
   "depth": "comprehensive",
   "report_style": "grok",
   "tools": ["web_search","pdf_parse","semantic_search"]
 }'

Typical JSON response schema (task ready/async)

{
  "task_id": "abc123",
  "status": "queued",
  "created_at": "2026-04-10T12:00:00Z",
  "poll_url": "/v1/tasks/abc123"
}

Result structure (when complete)

{
  "task_id": "abc123",
  "status": "completed",
  "report": {
    "summary": "...",
    "key_findings": ["..."],
    "evidence": [
      {"source":"https://arxiv.org/...", "snippet":"...", "offset": 102, "confidence": 0.92}
    ],
    "methodology": "...",
    "sources": [{"title":"...", "url":"...", "extracted_at":"..."}],
    "next_steps": ["..."]
  }
}

Use Cases

• Academic literature review: collect recent papers on a topic, extract methods and benchmarks, and create a synthesized annotated bibliography with DOIs and key quotes.
• Competitive product analysis: gather product pages, reviews, and changelogs, then produce a findings report ranking evidence for claims and feature differences.
• Regulatory or compliance research: aggregate statutes, guidance documents, and legal opinions; create an evidence-backed summary with citations and actionable gaps.
• News verification and fact-checking: pull articles across outlets, extract claims and timestamps, and surface corroborating or contradictory sources.
• Dataset curation: locate and extract data snippets from PDFs and web pages, normalize and index them into a vector store for downstream retrieval.

Getting Help / Contributing

See the repository on GitHub for source, issues, and contribution guidelines: https://github.com/reading-plus-ai/mcp-server-deep-research

For development, check the /docs directory in the repo for detailed tool interfaces, schema definitions, and examples of connecting agents (LLMs) to the MCP endpoint.