mcp-sequential-research 1.0.0

package/examples/example_workflow.md

@@ -0,0 +1,464 @@
+# Example Workflow: Photonic Computing Patent Research
+
+This example demonstrates the complete sequential research workflow for investigating photonic computing technology for AI acceleration.
+
+## Scenario
+
+**Research Goal**: Investigate the current state of photonic computing for neural network inference, identify key patents, and produce a citable research report.
+
+---
+
+## Step 1: Generate Research Plan
+
+### Request
+
+```json
+{
+  "method": "tools/call",
+  "params": {
+    "name": "sequential_research_plan",
+    "arguments": {
+      "topic": "photonic computing for neural network inference acceleration",
+      "depth": "standard",
+      "focus_areas": [
+        "silicon photonics integration",
+        "optical matrix multiplication",
+        "energy efficiency vs GPUs"
+      ],
+      "constraints": [
+        "patent-focused",
+        "post-2020 priority",
+        "english only"
+      ],
+      "output_format": "markdown"
+    }
+  }
+}
+```
+
+### Response (Abbreviated)
+
+```json
+{
+  "plan_id": "plan-abc123",
+  "topic": "photonic computing for neural network inference acceleration",
+  "created_at": "2025-01-13T10:00:00.000Z",
+  "summary": "Research plan with 8 queries at standard depth. Focus areas: silicon photonics integration, optical matrix multiplication, energy efficiency vs GPUs. Execution in 3 phases.",
+  "queries": [
+    {
+      "query_id": "q1",
+      "query_text": "What is photonic computing for neural network inference acceleration",
+      "query_type": "definition",
+      "extraction_goals": ["Primary definition", "Key characteristics", "Common examples", "Related terminology"],
+      "priority": "critical"
+    },
+    {
+      "query_id": "q2",
+      "query_text": "Current state of photonic computing for neural network inference acceleration",
+      "query_type": "current_state",
+      "extraction_goals": ["Current status", "Recent developments", "Key players or stakeholders", "Trends"],
+      "priority": "critical",
+      "depends_on": ["q1"]
+    },
+    {
+      "query_id": "q3",
+      "query_text": "Compare photonic computing for neural network inference acceleration",
+      "query_type": "comparison",
+      "extraction_goals": ["Key similarities", "Key differences", "Trade-offs", "Use case recommendations"],
+      "priority": "medium",
+      "depends_on": ["q1"]
+    },
+    {
+      "query_id": "q4",
+      "query_text": "photonic computing for neural network inference acceleration: silicon photonics integration",
+      "query_type": "current_state",
+      "priority": "high",
+      "depends_on": ["q1"]
+    },
+    {
+      "query_id": "q5",
+      "query_text": "photonic computing for neural network inference acceleration: optical matrix multiplication",
+      "query_type": "current_state",
+      "priority": "high",
+      "depends_on": ["q1"]
+    },
+    {
+      "query_id": "q6",
+      "query_text": "photonic computing for neural network inference acceleration: energy efficiency vs GPUs",
+      "query_type": "current_state",
+      "priority": "high",
+      "depends_on": ["q1"]
+    }
+  ],
+  "execution_order": [
+    ["q1"],
+    ["q2", "q3", "q4", "q5", "q6"]
+  ],
+  "estimated_sources": 18
+}
+```
+
+---
+
+## Step 2: Execute Queries
+
+Execute queries using appropriate MCP tools. Here's how each query maps to data sources:
+
+| Query | Source | MCP Tool |
+|-------|--------|----------|
+| q1 (definition) | Google Patents + Web | `search_patents`, Google Custom Search |
+| q2 (current_state) | News, Patents | Google Custom Search, `search_patents` |
+| q3 (comparison) | Technical papers | Google Custom Search |
+| q4-q6 (focus areas) | Patents primarily | `search_patents` |
+
+### Example: Query q1 via Google Patents MCP
+
+```json
+{
+  "method": "tools/call",
+  "params": {
+    "name": "search_patents",
+    "arguments": {
+      "query": "photonic computing neural network inference",
+      "num_results": 5,
+      "country": "US",
+      "after": "2020-01-01"
+    }
+  }
+}
+```
+
+### Example: Query q1 via Google Search MCP
+
+```json
+{
+  "method": "tools/call",
+  "params": {
+    "name": "google_custom_search",
+    "arguments": {
+      "query": "what is photonic computing for neural networks",
+      "num": 5
+    }
+  }
+}
+```
+
+---
+
+## Step 3: Normalize Results
+
+Transform raw API responses into the standard `raw_results` format:
+
+```json
+[
+  {
+    "query_id": "q1",
+    "success": true,
+    "data": {
+      "definition": "Photonic computing for neural network inference uses optical signals to perform matrix operations, achieving orders of magnitude lower latency and power consumption compared to electronic processors.",
+      "characteristics": [
+        "Parallel matrix multiplication via light interference",
+        "Near-zero energy for passive optical operations",
+        "Wavelength division multiplexing for high throughput",
+        "Femtosecond-scale operation speed"
+      ],
+      "examples": [
+        "Lightmatter Envise photonic chip",
+        "Intel Silicon Photonics accelerator",
+        "MIT photonic neural network prototype"
+      ]
+    },
+    "sources": [
+      {
+        "id": "S1",
+        "source_type": "document",
+        "title": "US11544545B2 - Photonic neural network system",
+        "url": "https://patents.google.com/patent/US11544545B2",
+        "accessed_date": "2025-01-13",
+        "excerpt": "A photonic neural network system comprising an array of Mach-Zehnder interferometers configured to perform matrix-vector multiplication..."
+      },
+      {
+        "id": "S2",
+        "source_type": "document",
+        "title": "US20230004850A1 - Optical inference accelerator",
+        "url": "https://patents.google.com/patent/US20230004850A1",
+        "accessed_date": "2025-01-13",
+        "excerpt": "An optical inference accelerator for deep learning workloads..."
+      },
+      {
+        "id": "S3",
+        "source_type": "web",
+        "title": "Photonic Computing Explained - MIT Technology Review",
+        "url": "https://technologyreview.com/photonic-computing-explained",
+        "accessed_date": "2025-01-13"
+      }
+    ]
+  },
+  {
+    "query_id": "q2",
+    "success": true,
+    "data": {
+      "current_status": "Photonic computing for AI inference is in early commercialization stage, with several startups shipping development kits and major tech companies investing in R&D.",
+      "recent_developments": [
+        "Lightmatter raised $154M Series C (2023)",
+        "Intel demonstrated 100Gbps silicon photonics (2024)",
+        "NVIDIA announced photonic interconnect research program"
+      ],
+      "key_players": [
+        "Lightmatter (Boston)",
+        "Luminous Computing (California)",
+        "Intel Silicon Photonics",
+        "Ayar Labs",
+        "MIT Photonics Research Lab"
+      ]
+    },
+    "sources": [
+      {
+        "id": "S4",
+        "source_type": "document",
+        "title": "US11675187B2 - Scalable photonic computing architecture",
+        "url": "https://patents.google.com/patent/US11675187B2",
+        "accessed_date": "2025-01-13"
+      },
+      {
+        "id": "S5",
+        "source_type": "web",
+        "title": "Lightmatter Series C Announcement",
+        "url": "https://lightmatter.co/news/series-c",
+        "accessed_date": "2025-01-13"
+      }
+    ]
+  },
+  {
+    "query_id": "q3",
+    "success": true,
+    "data": {
+      "similarities": [
+        "Both photonic and electronic accelerators target matrix operations",
+        "Both require custom software stacks",
+        "Both face manufacturing scaling challenges"
+      ],
+      "differences": [
+        "Photonic: passive operations consume near-zero energy",
+        "Electronic: mature ecosystem and tooling",
+        "Photonic: fundamentally analog computation",
+        "Electronic: digital precision but higher power"
+      ],
+      "recommendation": "Photonic accelerators are best suited for inference workloads with high matrix dimensions where latency and power are critical. Electronic accelerators remain preferred for training and general-purpose computing."
+    },
+    "sources": [
+      {
+        "id": "S6",
+        "source_type": "web",
+        "title": "Photonic vs Electronic AI Accelerators - IEEE Spectrum",
+        "url": "https://spectrum.ieee.org/photonic-vs-electronic-ai",
+        "accessed_date": "2025-01-13"
+      }
+    ]
+  },
+  {
+    "query_id": "q4",
+    "success": true,
+    "data": {
+      "current_status": "Silicon photonics integration enables manufacturing photonic chips using existing CMOS fabs, dramatically reducing costs.",
+      "recent_developments": [
+        "GlobalFoundries offering silicon photonics PDK",
+        "TSMC announced photonics integration roadmap",
+        "Intel shipping Ponte Vecchio with silicon photonics"
+      ],
+      "key_players": [
+        "Intel",
+        "GlobalFoundries",
+        "TSMC",
+        "Tower Semiconductor"
+      ]
+    },
+    "sources": [
+      {
+        "id": "S7",
+        "source_type": "document",
+        "title": "US11550101B2 - Silicon photonic integration method",
+        "url": "https://patents.google.com/patent/US11550101B2",
+        "accessed_date": "2025-01-13"
+      }
+    ]
+  },
+  {
+    "query_id": "q5",
+    "success": true,
+    "data": {
+      "current_status": "Optical matrix multiplication is the core operation enabling photonic neural networks, using interference patterns to compute dot products.",
+      "recent_developments": [
+        "Demonstrated 97% accuracy on ImageNet classification",
+        "10 TOPS/W efficiency achieved in lab conditions",
+        "Wavelength-multiplexed systems reaching 64 parallel channels"
+      ],
+      "key_players": [
+        "Lightmatter",
+        "MIT",
+        "Stanford Photonics Lab"
+      ]
+    },
+    "sources": [
+      {
+        "id": "S8",
+        "source_type": "document",
+        "title": "US11385449B2 - Optical matrix multiplication apparatus",
+        "url": "https://patents.google.com/patent/US11385449B2",
+        "accessed_date": "2025-01-13"
+      }
+    ]
+  },
+  {
+    "query_id": "q6",
+    "success": true,
+    "data": {
+      "current_status": "Photonic accelerators demonstrate 10-100x better energy efficiency than GPUs for inference, measured in TOPS/W.",
+      "recent_developments": [
+        "Lightmatter claims 1000 TOPS/W theoretical peak",
+        "NVIDIA H100 achieves ~10 TOPS/W for inference",
+        "Academic prototypes reaching 100 TOPS/W"
+      ],
+      "key_players": [
+        "Lightmatter",
+        "NVIDIA (benchmark baseline)",
+        "Luminous Computing"
+      ]
+    },
+    "sources": [
+      {
+        "id": "S9",
+        "source_type": "web",
+        "title": "Energy Efficiency Comparison: Photonic vs GPU",
+        "url": "https://example.com/photonic-gpu-efficiency",
+        "accessed_date": "2025-01-13"
+      }
+    ]
+  }
+]
+```
+
+---
+
+## Step 4: Compile Report
+
+### Request
+
+```json
+{
+  "method": "tools/call",
+  "params": {
+    "name": "sequential_research_compile",
+    "arguments": {
+      "plan": { /* full plan from step 1 */ },
+      "raw_results": [ /* array from step 3 */ ],
+      "include_sources": true,
+      "include_methodology": true,
+      "citation_style": "inline"
+    }
+  }
+}
+```
+
+### Response (markdown_report excerpt)
+
+```markdown
+# Research Report: photonic computing for neural network inference acceleration
+
+## Executive Summary
+
+This report presents research findings on **photonic computing for neural network inference acceleration**. The analysis covers 6 research queries across 5 sections. Key topics include: Overview, Current State, Comparison, Types and Categories.
+
+## Methodology
+
+This report was compiled using a sequential research approach:
+1. Research queries were generated based on the topic and focus areas
+2. Each query was executed to gather relevant information
+3. Results were compiled and organized into thematic sections
+4. Sources were consolidated and citations were verified
+
+## Overview
+
+Photonic computing for neural network inference uses optical signals to perform matrix operations, achieving orders of magnitude lower latency and power consumption compared to electronic processors. [1], [2]
+
+**Key Characteristics:**
+- Parallel matrix multiplication via light interference
+- Near-zero energy for passive optical operations
+- Wavelength division multiplexing for high throughput
+- Femtosecond-scale operation speed
+
+**Examples:**
+- Lightmatter Envise photonic chip
+- Intel Silicon Photonics accelerator
+- MIT photonic neural network prototype
+
+## Current State
+
+Photonic computing for AI inference is in early commercialization stage, with several startups shipping development kits and major tech companies investing in R&D. [4], [5]
+
+**Recent Developments:**
+- Lightmatter raised $154M Series C (2023)
+- Intel demonstrated 100Gbps silicon photonics (2024)
+- NVIDIA announced photonic interconnect research program
+
+**Key Players:**
+- Lightmatter (Boston)
+- Luminous Computing (California)
+- Intel Silicon Photonics
+- Ayar Labs
+- MIT Photonics Research Lab
+
+## Comparison
+
+**Similarities:**
+- Both photonic and electronic accelerators target matrix operations
+- Both require custom software stacks
+- Both face manufacturing scaling challenges
+
+**Differences:**
+- Photonic: passive operations consume near-zero energy
+- Electronic: mature ecosystem and tooling
+- Photonic: fundamentally analog computation
+- Electronic: digital precision but higher power
+
+**Recommendation:** Photonic accelerators are best suited for inference workloads with high matrix dimensions where latency and power are critical. Electronic accelerators remain preferred for training and general-purpose computing. [6]
+
+---
+
+### References
+
+- **[1]**: US11544545B2 - Photonic neural network system <https://patents.google.com/patent/US11544545B2>
+- **[2]**: US20230004850A1 - Optical inference accelerator <https://patents.google.com/patent/US20230004850A1>
+- **[3]**: Photonic Computing Explained - MIT Technology Review <https://technologyreview.com/photonic-computing-explained> (web)
+- **[4]**: US11675187B2 - Scalable photonic computing architecture <https://patents.google.com/patent/US11675187B2>
+- **[5]**: Lightmatter Series C Announcement <https://lightmatter.co/news/series-c> (web)
+- **[6]**: Photonic vs Electronic AI Accelerators - IEEE Spectrum <https://spectrum.ieee.org/photonic-vs-electronic-ai> (web)
+- **[7]**: US11550101B2 - Silicon photonic integration method <https://patents.google.com/patent/US11550101B2>
+- **[8]**: US11385449B2 - Optical matrix multiplication apparatus <https://patents.google.com/patent/US11385449B2>
+- **[9]**: Energy Efficiency Comparison: Photonic vs GPU <https://example.com/photonic-gpu-efficiency> (web)
+```
+
+---
+
+## Step 5: Save Outputs
+
+```bash
+# Create output directory
+mkdir -p research/photonic-computing-inference
+
+# Save files (in Claude Code, use Write tool)
+# research/photonic-computing-inference/report.md     <- markdown_report
+# research/photonic-computing-inference/sources.json  <- sources array
+# research/photonic-computing-inference/plan.json     <- original plan
+# research/photonic-computing-inference/raw_results.json <- normalized results
+```
+
+---
+
+## Key Takeaways
+
+1. **Plan first** — The structured plan ensures comprehensive coverage
+2. **Parallel execution** — Queries in the same execution group run concurrently
+3. **Normalize early** — Convert to standard schema immediately after each query
+4. **Stable citations** — `[S#]` format enables downstream claim-mining
+5. **Save everything** — Raw results allow re-processing without re-querying

package/package.json

@@ -0,0 +1,38 @@
+{
+  "name": "mcp-sequential-research",
+  "version": "1.0.0",
+  "description": "MCP server for sequential research planning and compilation",
+  "type": "module",
+  "main": "dist/index.js",
+  "types": "dist/index.d.ts",
+  "bin": {
+    "mcp-sequential-research": "./dist/index.js"
+  },
+  "scripts": {
+    "build": "tsc",
+    "dev": "tsx src/index.ts",
+    "lint": "eslint src --ext .ts",
+    "test": "echo \"Tests not yet implemented\" && exit 0"
+  },
+  "keywords": [
+    "mcp",
+    "model-context-protocol",
+    "research",
+    "sequential-thinking"
+  ],
+  "author": "",
+  "license": "MIT",
+  "dependencies": {
+    "@modelcontextprotocol/sdk": "^1.0.0",
+    "zod": "^3.23.0"
+  },
+  "devDependencies": {
+    "@types/node": "^20.11.0",
+    "eslint": "^8.57.0",
+    "tsx": "^4.7.0",
+    "typescript": "^5.3.0"
+  },
+  "engines": {
+    "node": ">=18.0.0"
+  }
+}