koncepto

Semantic concept graph MCP server for codebases — what your code means, not just what it does.

What

Code graphs (Aider repomap, GitNexus, Sourcegraph) capture structural relations: who imports who, who calls who. They miss semantic invariants — the cross-cutting concepts that live in code not related by imports:

• "Fix B" lives in 7 files but isn't a function or a class
• "All UI counting workload must exclude manual-override participants"
• "Sector value strings must match SectorAssignment.sector keys exactly"

koncepto is the curated semantic layer. Concepts in YAML, queryable via MCP tools, read at Step 0 before editing.

Status

Pre-alpha (v0.1.0-alpha.3 on npm). Schema and tool surface may break before 0.1.0 final. See roadmap.

Dogfooded against this repo itself: 5 concepts in .koncept/concepts/ cover the schema, the registry, the MCP tool contract, the monorepo shape, and the kebab-id naming convention. pnpm dogfood = koncepto verify against its own registry.

Quickstart

# Install in your project
pnpm add -D @yourtechtribe-labs/koncept-cli@alpha

# Bootstrap
npx koncepto init

# Write a concept (YAML)
$EDITOR .koncept/concepts/my-concept.yaml

# Verify
npx koncepto verify

# Register MCP server (Claude Code)
claude mcp add --scope user koncepto -- \
  npx -y @yourtechtribe-labs/koncept-mcp-server@alpha "$PWD"

Architecture

3 packages under pnpm workspace:

• @yourtechtribe-labs/koncept-core — Zod schema, YAML parser, indexer
• @yourtechtribe-labs/koncept-mcp-server — MCP stdio server (4 tools)
• @yourtechtribe-labs/koncept-cli — init, verify, list, link

License

MIT — see LICENSE.