Open SourceMIT License

Prelude

Prelude reads a codebase or a set of documents and produces a structured, machine-readable account of what is in it: what technologies are present, how the work is organized, what decisions were made, and what has changed. Ground truth that agents, diligence, and intelligence pipelines can rely on.

npm install View on GitHub

# Extract ground truth

$ prelude init

# Emit structured context

$ prelude export

# Run as MCP server

$ prelude serve

Ready for agents.

The input layer

Before you can analyze a codebase or a document set, you need to know what is actually in it. Prelude scans a repository, a folder, or a collection of documents and emits structured output: the technologies present, how the work is organized, the constraints that apply, the decisions that were made, and what has changed over time.

Diligence relies on it, intelligence pipelines ingest it, and agents reason over it. Structured input makes the analysis that follows something you can check rather than guess at. The core is open source and MIT licensed.

What Prelude extracts

Technology stack

Language, runtime, frameworks, database, ORM. Auto-detected from package files and configuration.

Architecture

Project structure, component patterns, key directories, monorepo layout, and how the codebase is organized.

Dependencies

What the asset depends on. Direct and transitive. The supply chain that underpins the work.

Decisions

Architecture Decision Records. Why the team chose one approach over another. The reasoning behind the code.

Constraints

Must-use technologies, patterns, and policies. The rules the team follows and why.

Changes

Timeline of modifications. What changed, when, and what the trajectory looks like over time.

Language support

JavaScript / TypeScript

package.json, frameworks, build tools, testing, styling

Python

pyproject.toml, requirements.txt, FastAPI, Django, Flask

Rust

Cargo.toml, dependencies, workspace detection

Go

go.mod, dependencies, module detection

Where Prelude fits

Agent grounding

Give a coding agent accurate context on your stack, architecture, and constraints, so it builds on what is actually there rather than guessing.

Technical diligence

What is actually in this codebase or document set? Prelude answers with structured data rather than opinions.

Portfolio monitoring

Track what changed across the codebases and documents you follow. Structured output that intelligence pipelines can ingest over time.

Attested lineage

Prelude output becomes part of the record: what was in the codebase or document at the time of analysis, preserved so it can be checked later.

Core commands

prelude init

Scans the repository and creates a .context/ directory with structured data: project metadata, stack, architecture, constraints, decisions, and changelog. Use --from-claude-md to bootstrap from an existing CLAUDE.md.

prelude export

Emits a structured document optimized for agent consumption. Supports --format claude-md, --format cursorrules, and --format json for different tool integrations.

prelude update

Re-scans and updates the context. Preserves manual annotations while refreshing auto-detected data. Edited fields are never overwritten.

prelude query

Scoped context lookup with topic, directory, and type filtering. Returns only the relevant slice of your codebase context.

prelude validate

Validates all context files against the Prelude Specification JSON schemas. Catches drift and malformed entries.

prelude decision

Records an architecture decision with rationale. Captures the reasoning behind technical choices for future analysis.

prelude serve

Runs Prelude as an MCP server. Exposes 3 tools (query, compact, status) and 3 resources for direct agent integration.

prelude watch

Monitors file changes in real time. Tracks development activity for session continuity and change analysis.

MCP server

Prelude runs as a Model Context Protocol server, giving AI agents direct access to your codebase context. Claude Code, Claude Desktop, and Cursor can query your stack, architecture, and constraints without manual copy-paste.

Tools

prelude_query - filtered context by topic, scope, type

prelude_compact - ~800 token context summary

prelude_status - which context files exist

Resources

prelude://context/full - complete context

prelude://context/compact - token-efficient summary

prelude://context/{type} - individual files

Generate config for your editor: prelude mcp-config

Ground truth starts here.

Open source and MIT licensed. The tool that turns a codebase or document set into structured intelligence agents can act on.

npm install -g prelude-context View Source