mcp-sequential-research

A Model Context Protocol (MCP) server for structured, reproducible research planning and report compilation. Designed for patent prior art research, technical investigations, and systematic literature reviews.

Features

• Structured Research Plans — Generate ordered queries with dependencies, priorities, and extraction goals
• Patent-Grade Analysis — Specialized query families for prior art research (broad, synonyms, competitors, limitations)
• Reproducible Workflows — Deterministic planning without external API calls
• Machine-Parseable Citations — Stable [S#] format for downstream claim-mining
• Prior Art Clustering — Automatic grouping of related sources
• Claim Risk Analysis — Flags for potential novelty issues
• Zod Validation — Full TypeScript type safety with runtime validation

Installation

npm install mcp-sequential-research

Or run directly:

npx mcp-sequential-research

Quick Start

Add to Claude Desktop

Add to your Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "sequential-research": {
      "command": "npx",
      "args": ["mcp-sequential-research"]
    }
  }
}

Add to Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "sequential-research": {
      "command": "npx",
      "args": ["mcp-sequential-research"]
    }
  }
}

Tools

sequential_research_plan

Generates a structured research plan with queries, extraction goals, and expected result schemas.

When to use:

• Multi-concept research topics
• Prior art investigations
• Systematic literature reviews
• When you need reproducible query coverage

Input:

{
  "topic": "photonic computing for neural network inference",
  "depth": "standard",
  "focus_areas": ["silicon photonics", "energy efficiency"],
  "constraints": ["patent-focused", "post-2020"],
  "output_format": "markdown"
}

Output:

• Ordered queries with priorities and dependencies
• Extraction goals per query
• Expected result schemas
• Execution order (parallel groups)
• Estimated source count

sequential_research_compile

Compiles raw research results into a structured report with citations and analysis.

When to use:

• After executing searches via Google Patents, web search, etc.
• When you need a citable report
• For prior art analysis with risk assessment

Input:

{
  "plan": { /* plan from sequential_research_plan */ },
  "raw_results": [
    {
      "query_id": "q1",
      "success": true,
      "data": { /* extracted data */ },
      "sources": [
        {
          "id": "S1",
          "source_type": "document",
          "title": "US11123456B2 - Photonic Processor",
          "url": "https://patents.google.com/patent/US11123456B2"
        }
      ]
    }
  ],
  "include_sources": true,
  "citation_style": "inline"
}

Output:

• Executive summary
• Organized report sections
• Inline citations [1], [2]
• Prior art clusters
• Claim risk flags
• Novelty gap suggestions
• Consolidated sources table

Patent-Grade Research

This server includes specialized query families for patent prior art research:

Patent Query Types

• Broad concept — High-level technology area
• Synonyms — Alternative terminology
• Problem/benefit framing — Problem-solution pairs
• Competitor/assignee — Known players in the space
• Component-level limitations — Narrow novelty claims

Web Query Types

• Academic — .edu sites, PDF papers
• Vendor documentation — Technical specifications
• Open source — GitHub, GitLab repositories
• Conference papers — IEEE, ACM, arXiv

Analysis Features

• Prior art clusters — Grouped related sources
• Claim risk flags — Potential novelty blockers
• Novelty gaps — Suggestions for differentiation

Workflow Example

// 1. Generate plan
const plan = await callTool("sequential_research_plan", {
  topic: "optical matrix multiplication",
  depth: "standard",
  focus_areas: ["photonic tensor cores"]
});

// 2. Execute queries using appropriate MCP tools
const results = [];
for (const query of plan.queries) {
  // Use Google Patents MCP, Google Search MCP, etc.
  const result = await executeQuery(query);
  results.push(normalizeResult(query.query_id, result));
}

// 3. Compile report
const report = await callTool("sequential_research_compile", {
  plan,
  raw_results: results,
  include_sources: true
});

// 4. Save outputs
await saveFile(`research/${slug}/report.md`, report.markdown_report);
await saveFile(`research/${slug}/sources.json`, report.sources);

Development

# Install dependencies
npm install

# Run in development mode
npm run dev

# Build for production
npm run build

# Run linter
npm run lint

API Reference

See docs/TOOL_CONTRACTS.md for complete JSON schemas and test data.

See docs/MCP_GUIDANCE.md for workflow guidance and best practices.

Citation Format

All citations use stable, machine-parseable identifiers:

Photonic computing achieves 10x efficiency [1], [2].

---
### References

| ID | Title | URL |
|----|-------|-----|
| [1] | US11123456B2 - Photonic Processor | https://patents.google.com/... |
| [2] | Silicon Photonics Overview | https://example.com/... |

This format is designed for downstream claim-mining tools.

Integration

Works with other MCP servers:

| Source Type | Recommended MCP | Tool | |-------------|-----------------|------| | Patents | Google Patents MCP | search_patents | | Web Search | Google Search MCP | google_search | | Web Scraping | Google Search MCP | read_webpage | | Memory | Memory MCP | search_nodes | | Academic | Semantic Scholar API | — |

License

MIT

Contributing

Contributions welcome! Please read CONTRIBUTING.md for guidelines.

Related

• Model Context Protocol
• @modelcontextprotocol/sdk
• Claude Code