dictionary
One name, everywhere
A workspace-wide dictionary of concepts. Rename an entry once and it propagates to every model that uses it — with "used in N models" rollups so you know the blast radius.
Concept modeling · humans + agents
The whiteboard where your AI agent holds a pen.
Koncept is a concept-modeling and architecture-diagramming tool with a built-in MCP server. Point Claude — or any MCP client — at your workspace and watch it draw. Every edit lands live on the canvas your team already has open.
Free plan · no card · MCP included on every tier
Your agent gets 32 tools,
not a screenshot.
Agents write through the same real-time sync log as the canvas — an agent edit and a human edit are indistinguishable to the system. Mint a workspace-scoped API key, give your agent a persona, and it becomes a teammate that models, reviews, and comments.
claude mcp add --transport http koncept \
https://<your-deployment>.convex.site/mcp \
--header "Authorization: Bearer <api-key>"- list_concepts
- create_node
- connect
- bulk_apply
- auto_layout
- render_concept
- validate_concept
- post_comment
- create_scenario
- search_dictionary
- +22 more
Draws like a peer
No import/export dance. The MCP server writes to the same Yjs update log the canvas reads, so agent strokes appear on human screens the moment they happen.
Sees its own work
render_concept returns a real PNG of the canvas — the same pixels you see. A multimodal agent draws, looks, and fixes its layout before you ever open the model.
Lints before you look
validate_concept catches dangling edges, overlaps, crossings, and undefined concepts — structured feedback agents act on without a human in the loop.
Everything a shared vocabulary needs.
Concept models keep a team honest about what words mean. Koncept gives the vocabulary a home, a review process, and concrete examples.
scenarios
Ground the abstract
Attach real cases to a model — "ACME's order #1042" — with example values per concept. Scenarios live beside the diagram, so real data never pollutes the picture. They even work on frozen models.
architecture views
Systems, not just concepts
A second model type with a seven-shape catalog — databases, external systems, actors, decisions — plus z-ordering. For service maps and data-flow drawings in the same workspace.
lifecycle
Draft → review → validated
Put models and dictionary entries through review with picked reviewers and approval thresholds. Optionally require branch-based "pull requests" for changes to validated models.
real-time
Multiplayer by default
Live co-editing over a Yjs log on Convex. Humans and agents share one edit stream — everyone sees everything as it happens, cursors to comments.
reviews & comments
Discuss on the diagram
Threads anchor to a model, a node, or an edge. Agents post and resolve comments through the same API — a "Claude reviewer" persona is one API-key option away.
Simple plans, honest caps.
Every plan — including Free — ships the full canvas, architecture views, scenarios, and the MCP server.
Free
$0
- 3 concept models
- Solo workspace
- MCP server + API keys
- Scenarios & architecture views
grows with you
Team
One plan, three sizes — step up as the team grows.
members
$12/mo
- 15 models · 5 members
- Workspace dictionary
- Comments & review flows
- Validation lifecycle & branch "PRs"
- Dictionary domains — from Team 20
Enterprise
Custom
- Unlimited models & members
- Everything in Team · 50
- Talk to us about your needs
Prices in USD per workspace, per month. Cancel anytime.