Two ways to use AI

Tesseract offers two distinct AI interfaces, each suited to different tasks:

Use the Console when you want to quickly sketch an architecture from scratch or ask questions about the current diagram. Switch to Claude Code when you need to scan your project's source code and build the diagram from real files.

Claude Console

The Claude Console is a collapsible chat panel in the bottom-left corner of the editor. It gives Claude direct access to your architecture graph through MCP tools, so it can add components, create connections, navigate subgraphs, and manage flows — all from a conversation.

Connecting

If the Console shows a Sign in button, you need to authenticate with your Anthropic account:

1 Launch Claude Code

Open a terminal and run claude to start Claude Code. If you don't have it yet, install it from claude.ai/code.

2 Log in to your Anthropic account

Follow the prompts in Claude Code to authenticate. This stores credentials locally that Tesseract can use.

3 Click Sign in

Back in Tesseract, click Sign in. The Console detects your Claude Code credentials and connects automatically.

What the Console can do

What the Console cannot do

The Console does not have access to your source code. It cannot read files, run commands, or analyze your codebase. It only sees the architecture graph inside Tesseract. For code-aware features, use Claude Code with the MCP connection described below.

Claude Code & MCP

Claude Code is Anthropic's CLI agent. When connected to Tesseract via MCP (Model Context Protocol), it can read your source code and modify your architecture diagram at the same time — bridging the gap between code and design.

Setting up the connection

Tesseract runs a local MCP server automatically when the application starts. You can find its address in the MCP → Config menu item (e.g. localhost:7440/mcp). Clicking this item copies the address to your clipboard.

The quickest way to register the server in Claude Code:

claude mcp add --transport http tesseract http://localhost:7440/mcp

Alternatively, you can create or edit a .mcp.json file at the root of your project:

{
  "mcpServers": {
    "tesseract": {
      "type": "http",
      "url":  "http://localhost:7440/mcp"
    }
  }
}

Then launch Claude Code with claude in your terminal. It discovers the MCP config and connects to Tesseract. You can now ask Claude to build your architecture from code.

Make sure Tesseract is running before starting Claude Code. The MCP server must be reachable at the address shown in MCP → Config.

Auto-approving MCP tools

By default, Claude Code asks for confirmation the first time it calls each MCP tool. To approve all Tesseract tools at once, add mcp__tesseract to your project's permission allow list:

// .claude/settings.json
{
  "permissions": {
    "allow": [
      "mcp__tesseract"
    ]
  }
}

This approves every tool exposed by the Tesseract MCP server. You only need to do this once per project. Without this, Claude Code will prompt for each tool on first use (you can approve them individually at that point).

Pre-built skills

Tesseract ships with five Claude Code skills that automate common architecture tasks. To install them, go to MCP → Install Skills and choose:

Once installed, each skill can scan your codebase and populate the diagram accordingly.

Skill What it does
arch-overview Scans your project structure (Docker, package managers, entry points, configs) and creates a high-level architecture diagram.
arch-services Finds services and APIs (routes, controllers, gRPC, GraphQL, OpenAPI) and adds them to the diagram.
arch-data Discovers the data layer (ORM models, migrations, database connections, caches, storage) and maps it out.
arch-infra Reads infrastructure configs (Kubernetes, Terraform, CI/CD pipelines, reverse proxies) and adds them.
arch-detail Zooms into a single component and creates a detailed subgraph of its internals (modules, routes, business logic).

To use a skill, ask Claude Code in the terminal. For example:

# Generate the high-level architecture
> Scan my project and create an architecture overview

# Detail a specific component
> Create a detailed subgraph for the API Gateway component

Claude Code reads your files, identifies the relevant pieces, and uses the MCP tools to populate Tesseract. You see the diagram update in real time.

MCP tools

The MCP server exposes a set of tools that Claude (Console or Code) uses to interact with the diagram. You don't call these directly — Claude uses them behind the scenes — but understanding them helps you craft better prompts.

Architecture tools

Tool Description
get_graph Read the architecture at the current level or a specific subgraph path.
list_types List all available component types from the library.
list_components List all components in the diagram.
add_component Add a new component with name, type, layer, tech, and description.
update_component Modify an existing component's properties.
remove_component Delete a component and its connections.
add_connection Create a connection between two components.
update_connection Modify a connection's protocol, description, curvature, or direction.
remove_connection Delete a connection.
annotate Add a note or annotation to a component.
screenshot Capture the current 3D view as a PNG image.
export_mermaid Export the architecture as a Mermaid flowchart (text).
import_mermaid Import a Mermaid flowchart and replace the current architecture.
update_project Set the project title and/or description.

Navigation tools

Tool Description
look_at Select a component or enter its subgraph to bring it to the user's attention.
get_user_context Get the current view path and selection state.

Flow tools

Tool Description
highlight_path Temporarily highlight a sequence of connections on the canvas.
clear_highlights Remove temporary highlights.
save_flow Save the current highlights as a persistent flow.
list_flows List all saved flows.
show_flow Display a saved flow on the canvas.
update_flow Modify a flow's label, paths, or segments.
delete_flow Remove a saved flow.
Flow tools are a Pro feature. See Data Flows for more details.

Layer tools

Tool Description
list_layers List all layers with their names and colors.
add_layer Add one or more layers (batch).
update_layer Rename or recolor layers, with automatic component reassignment.
remove_layer Remove layers (orphaned components move to the first remaining layer).
reorder_layers Reorder layers from top to bottom.
Layer tools are a Pro feature. See MCP API Reference for full parameter details.

MCP resources

In addition to tools, the MCP server exposes read-only resources that Claude can query for context:

Resource URI Description
architecture://overview High-level summary: component count, connection count, layers.
architecture://component/{id} Full details of a single component.
architecture://connections All connections with resolved component names.
architecture://layer/{name} All components belonging to a specific layer.
library://components Full component type catalog grouped by collection and category.
library://collection/{id} Components in a specific library collection.

Real-time sync

When Claude modifies the architecture (through either the Console or Claude Code), changes appear on your canvas in real time.

You and Claude can work on the diagram simultaneously. The locking system ensures you never overwrite each other's changes.

Tips for effective prompts

Previous
Save & Load