buildnext

Synthesize customer feedback into actionable product themes. CLI + MCP server.

Point buildnext at your interview transcripts, Canny board, Slack channels, or Airtable tables. It uses an LLM to find cross-cutting themes, cite exact quotes as evidence, and suggest what to build next.

Install

npm install -g buildnext

Quick Start

# Create a config file
buildnext init

# Set your API key
export ANTHROPIC_API_KEY=sk-ant-...

# Put interview transcripts in ./feedback/
# Run synthesis
buildnext run

What It Does

1. Ingests customer data from multiple sources (files, Canny, Slack, Airtable, Gong)
2. Synthesizes themes across all feedback using an LLM
3. Prioritizes themes into ranked feature candidates with effort/impact estimates
4. Generates specs with user stories, acceptance criteria, and data model suggestions
5. Breaks down tasks into implementation-ready work items for developers or coding agents

The full pipeline: customer feedback → themes → features → specs → dev tasks. Each step is independently useful. The output is markdown. Pipe it, save it, or read it in your terminal.

Data Sources

Source	What it pulls	Auth
Files	.md, .txt, .docx from a directory	None
Canny	Feature requests, votes, comments	API key
Slack	Messages from specific channels	Bot token
Airtable	Records from any table	Personal access token
Gong	Call transcripts	Access key + secret

Configuration

Create .buildnextrc.json in your project root:

{
  "llm": {
    "provider": "anthropic",
    "model": "claude-sonnet-4-20250514"
  },
  "sources": {
    "files": {
      "directory": "./feedback",
      "extensions": [".md", ".txt", ".docx"]
    },
    "canny": {
      "apiKey": "your-key",
      "boardIds": ["board-id"],
      "since": "2025-01-01"
    },
    "slack": {
      "botToken": "xoxb-your-token",
      "channels": ["C0123FEEDBACK"],
      "since": "2025-01-01"
    },
    "airtable": {
      "apiKey": "pat...",
      "baseId": "appXXX",
      "tableId": "tblXXX",
      "contentFieldId": "fldXXX",
      "titleFieldId": "fldXXX",
      "dateFieldId": "fldXXX"
    }
  },
  "output": {
    "maxThemes": 10,
    "minEvidence": 2
  }
}

API keys can also be set via environment variables: ANTHROPIC_API_KEY, OPENAI_API_KEY, CANNY_API_KEY, SLACK_BOT_TOKEN, AIRTABLE_API_KEY.

CLI Commands

# Synthesis (step 1)
buildnext run                        # Ingest feedback and synthesize themes
buildnext run --sources files,canny  # Only specific sources
buildnext run --format json          # JSON output
buildnext run --output report.md     # Write to file
buildnext run --dry-run --verbose    # Count items without calling LLM
buildnext run --no-cache             # Skip cache

# Prioritization (step 2)
buildnext prioritize                 # Rank themes into feature candidates
buildnext prioritize --max 5         # Top 5 features only
buildnext prioritize --context "B2B SaaS, usage-based billing"

# Spec generation (step 3)
buildnext spec                       # Spec for the top-ranked feature
buildnext spec --feature 2           # Spec for feature at index 2
buildnext spec --output spec.md      # Write to file

# Task breakdown (step 4)
buildnext tasks                      # Break down top feature into dev tasks
buildnext tasks --feature 1          # Tasks for feature at index 1
buildnext tasks --context "Next.js + NestJS monorepo"

# Utilities
buildnext init                       # Create config file
buildnext sources                    # List configured sources
buildnext cache clear                # Clear cached results
buildnext serve                      # Start MCP server

MCP Server

buildnext can run as an MCP server for Claude Code, Cursor, or any MCP client.

buildnext serve

Tools

• synthesize_feedback — Ingest feedback and synthesize themes
• list_sources — Show configured sources
• get_themes — Return themes from the last synthesis
• prioritize_features — Rank themes into feature candidates with scores
• generate_spec — Generate a product spec for a feature
• generate_tasks — Break down a spec into implementation tasks

Resource

• buildnext://report/latest — The most recent synthesis report

Claude Code Configuration

Add to your Claude Code MCP settings:

{
  "mcpServers": {
    "buildnext": {
      "command": "npx",
      "args": ["buildnext", "serve"]
    }
  }
}

Then ask: "What should we build next?" or "Prioritize our top features" or "Write a spec for the #1 feature and break it into tasks."

LLM Support

• Anthropic (Claude) — default, best for structured analysis
• OpenAI (GPT-4o) — set "provider": "openai" in config

Example Output

# Product Feedback Synthesis
Generated: 2026-03-31 | Model: claude-sonnet-4-20250514 | 70 items from files, canny

## Themes

### 1. Cost Visibility is the Top Pain Point (high urgency)
Customers consistently struggle to understand per-customer and per-feature
costs. They can see total spend but can't attribute it.

Mentioned in 12 items.

**Evidence:**
- "I have no idea which customer is actually profitable" — interview-sarah.md
- "The bill just says $4,000. I need to know WHY it's $4,000" — canny

**Suggested Actions:**
- Add per-customer cost breakdown to the dashboard
- Build a cost attribution report that maps usage to revenue

Full Pipeline Example

# Step 1: Synthesize themes from customer feedback
buildnext run --output themes.md

# Step 2: Prioritize into feature candidates
buildnext prioritize --context "B2B SaaS, usage-based billing" --output features.md

# Step 3: Generate spec for the top feature
buildnext spec --output spec.md

# Step 4: Break into dev tasks
buildnext tasks --context "Next.js frontend, NestJS backend" --output tasks.md

Or via MCP in Claude Code / Cursor — just ask naturally:

"Synthesize our customer feedback, prioritize features, write a spec for the top one, and break it into tasks."

The MCP server maintains pipeline state across tool calls, so each step feeds into the next automatically.

License

Apache 2.0