MCP Servers

模型上下文协议服务器、框架、SDK 和模板的综合目录。

C
Cameo MCP Bridge
作者 @ajhcs

MCP server bridging AI assistants to CATIA Magic / Cameo Systems Modeler for SysML/UML model creation, querying, and manipulation

创建于 3/3/2026
更新于 about 1 month ago
Repository documentation and setup instructions

Cameo MCP Bridge

An MCP (Model Context Protocol) server that connects AI coding assistants to CATIA Magic / Cameo Systems Modeler -- the industry-standard MBSE tool for SysML and UML modeling.

This lets Claude Code (or any MCP-compatible client) query, create, modify, and visualize SysML/UML models inside a running Cameo instance through 49 tools covering capability negotiation, methodology-aware OOSEM workflows, elements, relationships, native requirement matrices, diagrams, reusable verification, specifications, and Groovy macro execution.

Claude Code  <--stdio/MCP-->  Python MCP Server  <--HTTP/REST-->  Java Plugin (Cameo JVM)

Why This Exists

MBSE tools like Cameo are powerful but manual. With this bridge, an AI assistant can:

  • Build models from requirements -- "Create a state machine for ATM operations with idle, active, and maintenance states"
  • Query and navigate models -- "Show me all blocks with the <<requirement>> stereotype"
  • Generate diagrams -- Create sequence diagrams, BDDs, IBDs, state machines, and populate them with elements
  • Export diagram images -- Get PNG snapshots of any diagram as base64
  • Run Groovy scripts -- Escape hatch for anything the structured tools don't cover
  • Inspect and modify specifications -- Read/write any UML property, tagged value, or constraint

How Is This Different?

| Project | Approach | Status | |---------|----------|--------| | This project | Talks directly to Cameo's Java API via an embedded plugin | Production-tested | | SysML v2 API MCP Server | Connects to SysML v2 REST API (tool-agnostic) | Early stage, C++ | | EA MCP Server | Enterprise Architect integration | Closed-source, Windows-only | | Dassault's prototype | SysML v2 + MCP demo | Promotional, not shipped |

This is the only open-source MCP server that integrates directly with a running Cameo instance, giving full access to SysML v1 models and the complete Cameo API surface.

Architecture

+-------------------+         +---------------------+         +---------------------------+
|                   |  stdio  |                     |  HTTP   |                           |
|  Claude Code /    |-------->|  Python MCP Server  |-------->|  Java Plugin              |
|  Any MCP Client   |<--------|  (cameo_mcp)        |<--------|  (CameoMCPBridgePlugin)   |
|                   |   MCP   |                     | JSON    |                           |
+-------------------+         +---------------------+         +---------------------------+
                                                               |                         |
                                                               |  127.0.0.1:18740        |
                                                               |                         |
                                                               |  Handlers:              |
                                                               |  - ProjectHandler       |
                                                               |  - ElementQueryHandler  |
                                                               |  - ElementMutationHandler|
                                                               |  - RelationshipHandler  |
                                                               |  - MatrixHandler        |
                                                               |  - DiagramHandler       |
                                                               |  - ContainmentTreeHandler|
                                                               |  - SpecificationHandler |
                                                               |  - MacroHandler         |
                                                               +---------------------------+
                                                                         |
                                                               +---------v---------+
                                                               | CATIA Magic /     |
                                                               | Cameo JVM         |
                                                               | (OpenAPI, EMF,    |
                                                               |  SessionManager)  |
                                                               +-------------------+

Data flow for a write operation:

  1. MCP client calls a tool (e.g., cameo_create_element)
  2. Python server translates to HTTP POST to the Java plugin
  3. Java handler dispatches to Swing EDT via EdtDispatcher
  4. On EDT: opens a SessionManager session, executes the operation, closes the session
  5. JSON response flows back through the layers

All write operations are session-wrapped for undo/redo support. Read operations run on the HTTP thread pool (Cameo model reads are thread-safe).

Prerequisites

  • CATIA Magic / Cameo Systems Modeler 2024x or newer (any bundle: Systems of Systems Architect, Cyber Systems Engineer, etc.)
  • Java 17 JDK available to Gradle
  • Python 3.10+ with pip
  • Gradle 8.x (wrapper included)

Installation

Quick Install

git clone https://github.com/ajhcs/cameo-mcp-bridge.git
cd cameo-mcp-bridge

# Set your Cameo install path (default: D:/DevTools/CatiaMagic)
export CAMEO_HOME="/path/to/your/CatiaMagic"

# Optional: point the installer/Gradle at a Java 17 JDK explicitly
export JDK17_HOME="/path/to/jdk-17"

./install.sh

The install script:

  1. Builds the Java plugin with Gradle and passes CAMEO_HOME through automatically
  2. Deploys it to $CAMEO_HOME/plugins/com.claude.cameo.bridge/
  3. Creates or reuses mcp-server/.venv/ when not already inside a virtualenv
  4. Installs the Python MCP server into that environment
  5. Registers the MCP server with Claude Code when the claude CLI is available

Manual Install

1. Build the Java plugin:

cd plugin
./gradlew assemblePlugin -PcameoHome="/path/to/CatiaMagic" -Pjdk17Home="/path/to/jdk-17"

Gradle must run on a Java 17 JDK. You can also set JDK17_HOME or JAVA17_HOME instead of passing -Pjdk17Home=....

2. Deploy to Cameo:

Copy the contents of plugin/build/plugin-dist/com.claude.cameo.bridge/ to:

<CAMEO_HOME>/plugins/com.claude.cameo.bridge/

3. Install the Python server:

cd mcp-server
python3 -m venv .venv
. .venv/bin/activate
python -m pip install -e .

On Windows shells, use .venv\\Scripts\\activate instead.

4. Register with your MCP client:

For Claude Code:

claude mcp add cameo-bridge --scope user -- /absolute/path/to/mcp-server/.venv/bin/python -m cameo_mcp.server

On Windows, the interpreter path is typically .venv\\Scripts\\python.exe.

For other MCP clients, configure stdio transport with the venv interpreter and command -m cameo_mcp.server.

5. Restart CATIA Magic, open a project, and verify:

> Check cameo status

If a newly added MCP tool returns HTTP 404 after an update, the Python server and Java plugin are out of sync. Rebuild/redeploy the plugin, then restart CATIA Magic so the new HTTP handlers are loaded.

The Python side now performs a capability handshake against the plugin before non-status operations. If cameo_status or cameo_get_capabilities reports compatibility.clientCompatible = false, stop and redeploy the matching plugin before proceeding.

The Python MCP layer also ships a Phase 2 methodology surface for bounded OOSEM workflows. These tools build named artifact recipes, workflow guidance, conformance checks, and compact review packets on top of the low-level bridge.

Configuration

| Environment Variable | Default | Description | |---------------------|---------|-------------| | CAMEO_BRIDGE_PORT | 18740 | HTTP port for the bridge (must match both sides) | | JDK17_HOME | unset | Optional Java 17 home used by install.sh and Gradle | | JAVA17_HOME | unset | Alternate Java 17 home override | | CAMEO_MCP_STRUCTURED_RESPONSES | deprecated | Structured MCP object responses are now always used; this flag is kept only for backward compatibility with older docs |

The Java plugin reads the port from system property cameo.mcp.port (default 18740). To change it, add to your Cameo *.vmoptions file:

-Dcameo.mcp.port=18741

And set CAMEO_BRIDGE_PORT=18741 in your environment before launching Claude Code.

Tool Reference

Project & Session (5 tools)

| Tool | Description | |------|-------------| | cameo_status | Check plugin health and report client/plugin compatibility | | cameo_get_capabilities | Get machine-readable endpoint/capability metadata | | cameo_get_project | Get project name, file path, and root model ID | | cameo_save_project | Save the project to disk | | cameo_reset_session | Force-close a stuck editing session (recovery tool) |

Methodology Packs (6 tools)

| Tool | Description | |------|-------------| | cameo_list_methodology_packs | List built-in methodology packs such as oosem | | cameo_get_methodology_pack | Get one pack's phases, recipes, naming rules, and evidence structure | | cameo_get_methodology_guidance | Explain which artifact is missing next and why | | cameo_execute_methodology_recipe | Execute a bounded recipe and return receipts, conformance, and review packet output | | cameo_validate_methodology_recipe | Validate normalized artifact snapshots against recipe expectations | | cameo_generate_review_packet | Generate a compact evidence bundle and Markdown review packet without mutating the model |

Elements (8 tools)

| Tool | Description | |------|-------------| | cameo_query_elements | Search by type, name, package, stereotype, with paging and compact/full views | | cameo_get_element | Get full details of a single element | | cameo_create_element | Create a new model element | | cameo_modify_element | Change name or documentation | | cameo_delete_element | Remove an element and its children | | cameo_get_containment_tree | Browse the project hierarchy | | cameo_list_containment_children | Page/filter immediate children for large models with compact/full views | | cameo_apply_profile | Apply a profile to a model/package so custom stereotypes become usable |

Supported element types: Package, Profile, Stereotype, Class, Block, Property, FlowProperty, Port, Interface, Activity, UseCase, Actor, StateMachine, State, Pseudostate, InitialState, Requirement, InterfaceBlock, ConstraintBlock, ValueType, DataType, Signal, Enumeration, Component, Comment, Constraint, CallBehaviorAction, OpaqueAction, ActivityPartition, InitialNode, ActivityFinalNode, FlowFinalNode, DecisionNode, MergeNode, ForkNode, JoinNode, InputPin, OutputPin, Operation

SysML aliases such as Block, Requirement, ConstraintBlock, InterfaceBlock, ValueType, and FlowProperty rely on the SysML profile being available. The bridge now treats missing required stereotypes as an error instead of silently producing plain UML elements. Pseudostate currently maps to an initial pseudostate in the MVP.

For large projects, prefer cameo_list_containment_children over cameo_get_containment_tree. The recursive tree endpoint is still available for compatibility, but it can produce very large responses on real models.

Stereotypes & Tagged Values (3 tools)

| Tool | Description | |------|-------------| | cameo_apply_stereotype | Apply a stereotype (e.g., <<block>>, <<requirement>>) | | cameo_set_tagged_values | Set tagged values on a stereotyped element | | cameo_set_stereotype_metaclasses | Bind a stereotype to UML metaclasses using Cameo's supported API |

If you create a custom profile through MCP, the typical sequence is:

  1. Create the Profile
  2. Create Stereotype elements inside it, optionally with metaclasses=[...]
  3. Call cameo_apply_profile on the target model/package
  4. Apply the new stereotypes to model elements

Relationships (2 tools)

| Tool | Description | |------|-------------| | cameo_create_relationship | Create a relationship between two elements | | cameo_get_relationships | Query relationships for an element |

Supported relationship types: Association, DirectedAssociation, Composition, Generalization, Dependency, ControlFlow, ObjectFlow, Transition, Connector, InformationFlow, ItemFlow, Allocate, Satisfy, Derive, Refine, Trace, Verify, Include, Extend

Connector supports nested-port partWithPort ownership. InformationFlow and ItemFlow support structured realizingConnector, conveyed, and SysML itemProperty payload data for IBD item-flow workflows. When you pass an IBD context element as ownerId, the bridge resolves the actual InformationFlow/ItemFlow containment to the nearest package because Cameo does not allow those relationships to be owned directly by a block.

Matrices (4 tools)

| Tool | Description | |------|-------------| | cameo_list_matrix_kinds | List validated native matrix kinds, aliases, and example type domains | | cameo_list_matrices | List supported native requirement matrices in the project | | cameo_get_matrix | Read one native refine/derive matrix with rows, columns, and populated cells | | cameo_create_matrix | Create a native refine or derive requirement matrix artifact |

Supported matrix kinds: refine, derive

These tools target Cameo's native matrix artifacts:

  • refine -> Refine Requirement Matrix
  • derive -> Derive Requirement Matrix

This matrix family is separate from the diagram shape/path API. It manages native requirement-matrix artifacts and returns row/column/cell data directly.

cameo_create_matrix also accepts optional row_types and column_types lists so native refine matrices can target mission artifacts such as UseCase, Property, or SysML stereotypes instead of being fixed to Block. Use cameo_list_matrix_kinds to see the validated kind aliases and example type domains.

Diagrams (15 tools)

| Tool | Description | |------|-------------| | cameo_list_diagram_types | List validated diagram request tokens, aliases, and native Cameo types | | cameo_list_diagrams | List all diagrams in the project | | cameo_create_diagram | Create a new diagram (18 types supported) | | cameo_add_to_diagram | Place a model element on a diagram canvas and return its presentationId | | cameo_get_diagram_image | Export a diagram as base64-encoded PNG | | cameo_auto_layout | Apply Cameo's built-in auto-layout | | cameo_list_diagram_shapes | List all shapes/paths with presentation IDs, bounds, and counts | | cameo_get_shape_properties | Read the current display properties of one diagram shape | | cameo_move_shapes | Reposition/resize shapes on a diagram with per-item results | | cameo_delete_shapes | Remove shapes from a diagram (model elements preserved) | | cameo_add_diagram_paths | Draw relationship paths between shapes on a diagram | | cameo_set_shape_properties | Set display properties (colors, compartment visibility, etc.) with receipts | | cameo_set_shape_compartments | Apply normalized compartment visibility controls to one shape | | cameo_reparent_shapes | Move existing presentation elements under new container shapes | | cameo_route_paths | Update path breakpoints, endpoints, and label reset behavior |

Validated diagram request tokens: Class, Package, UseCase, Activity, Sequence, StateMachine, Component, Deployment, CompositeStructure, Object, Communication, InteractionOverview, Timing, Profile, BDD, IBD, Requirement Diagram, Parametric Diagram

Use cameo_list_diagram_types if you want the accepted aliases too. Common forms such as InternalBlockDiagram, SysML IBD, ClassDiagram, and StateMachineDiagram are normalized to the validated token set automatically.

Verification (2 tools)

| Tool | Description | |------|-------------| | cameo_verify_matrix_consistency | Check matrix row/column membership, populated-cell count, dependency names, and density | | cameo_verify_diagram_visual | Check rendered PNG validity, diagram/path presence, content coverage, and coarse overlap risk |

These verification tools are Python-side wrappers over the bridge's native diagram and matrix readback. They are intended for repeatable regression checks, visual sanity validation, and lightweight quantitative validation without dropping to raw macros.

Specification (3 tools)

| Tool | Description | |------|-------------| | cameo_get_specification | Read all UML properties, tagged values, and constraints | | cameo_set_specification | Write properties, tagged values, or constraint fields | | cameo_set_usecase_subject | Set or clear the UML subject classifier for a use case |

Macros (1 tool)

| Tool | Description | |------|-------------| | cameo_execute_macro | Execute arbitrary Groovy scripts inside the Cameo JVM |

The macro tool is an escape hatch for operations not covered by the structured tools. Scripts have full access to the Cameo OpenAPI, with project, application, primaryModel, and ef (ElementsFactory) pre-injected into the script context.

Important: Scripts that modify the model must manage their own sessions:

import com.nomagic.magicdraw.openapi.uml.SessionManager

SessionManager.getInstance().createSession(project, "My operation")
try {
    // ... modify model ...
    SessionManager.getInstance().closeSession(project)
} catch (Exception e) {
    SessionManager.getInstance().cancelSession(project)
    throw e
}

If a macro fails mid-session, use cameo_reset_session to recover.

Usage Examples

Create a SysML Block

Create a block called "Sensor" in the root model package

The AI will:

  1. Call cameo_get_project to find the root model ID
  2. Call cameo_create_element with type "Block", name "Sensor", and the root model ID as parent

Build A State Machine

Create a state machine for an ATM with states OFF, IDLE, ACTIVE, and MAINTENANCE.
Add an initial pseudostate and transitions: initial -> OFF, OFF -> IDLE on startup,
IDLE -> ACTIVE on card insert, ACTIVE -> IDLE on transaction complete, and any state
-> MAINTENANCE on service request.

Query and Modify

Find all requirements in the project and show me their IDs and text

Export Diagrams

Export the "System Overview" diagram as a PNG and save it to my desktop

Run a Groovy Script

Run a macro that lists all profiles loaded in the current project

Known Limitations

Diagram Layout (Primary Pain Point)

The bridge builds models correctly -- elements, relationships, directionality, stereotypes, and structure all come out right. The main gap is diagram presentation: layout, spacing, and visual properties of complex diagrams often need manual adjustment in Cameo's GUI.

cameo_list_diagram_shapes now discovers nested presentation elements recursively, but complex nested editing is still incomplete. Sequence diagrams, composite states, and other deeply nested presentations can still require manual cleanup or Groovy fallbacks. This means:

| What Doesn't Work | Why | |---|---| | Spacing messages vertically in sequence diagrams | Message arrows aren't top-level shapes; no Y-position control during creation | | Moving messages relative to combined fragments (ref boxes) | Same -- messages and fragments are nested presentation elements | | Self-messages (lifeline to itself) | PresentationElementsManager.createPathElement() fails when source == target | | Showing region names in composite states | Region labels are nested inside the state shape; can't be found or configured | | Resizing nested states to show full entry/exit behaviors | Sub-states in regions aren't accessible through the flat shape listing | | Controlling transition label text display | Transition paths inside composite states are nested |

Workarounds:

  • Use cameo_execute_macro with Groovy scripts that access nested presentation elements directly
  • Use cameo_auto_layout (works well for simple diagrams, less so for complex state machines)
  • For sequence diagrams: the model is correct, so manual drag-and-drop in Cameo takes 5-10 minutes
  • For state machines: widen shapes and toggle region name visibility manually

Current direction: Keep expanding structured editing for nested presentation elements so fewer workflows require macros.

Not Yet Implemented

  • Generic matrix/table artifact handler -- native matrix support currently covers refine/derive requirement matrices only
  • Remove stereotype -- can apply but not remove
  • Delete/rename diagrams -- diagrams can be created and populated but not deleted or renamed through the bridge
  • Element reparenting -- cannot move elements between packages
  • Undo/redo -- sessions support undo in Cameo's UI, but no MCP tool to trigger it
  • Bulk operations -- creating N elements requires N sequential API calls
  • Model change notifications -- purely request/response; no event subscription
  • File-based diagram export -- cameo_get_diagram_image returns base64 over the wire; no option to save directly to a file path (use cameo_execute_macro with ImageExporter.export(dpe, ImageExporter.PNG, file) as a workaround)

API Gaps

  • DurationConstraint / TimeConstraint creation through macros is unreliable due to complex ownership chains in the Cameo API (DurationInterval.setMin/setMax requires Duration instances with specific ownership that the API rejects); add these manually in Cameo's UI
  • Large diagram images may exceed MCP client token limits when returned as base64; use the macro workaround above to save directly to disk
  • format parameter on cameo_get_diagram_image is accepted but ignored (always exports PNG)
  • Session recovery edge case -- if a macro crashes mid-transaction, cameo_reset_session may itself throw TransactionAlreadyCommitedException; in this case, saving and reopening the project is the most reliable recovery

Compatibility

  • Tested with CATIA Magic Systems of Systems Architect 2024x
  • Should work with any Cameo Systems Modeler 2024x bundle (2024x+)
  • Requires Groovy script engine (bundled with Cameo) for macro execution
  • The Gradle build requires access to Cameo's lib/ directory for compile-time dependencies

Security Considerations

This bridge is designed for local development use only.

  • The HTTP server binds to 127.0.0.1 (localhost only) -- not accessible from the network
  • There is no authentication on the HTTP endpoints
  • The cameo_execute_macro tool executes arbitrary Groovy code inside the Cameo JVM with full access to the filesystem, network, and classloader
  • CORS headers are set to * (wildcard) -- any webpage in a local browser could theoretically make requests to the bridge

Do not expose the bridge port to the network, run it on shared/multi-user machines without additional access controls, or use it in production environments without adding authentication.

Project Structure

cameo-mcp-bridge/
  mcp-server/                          # Python MCP server
    cameo_mcp/
      __init__.py
      client.py                        # HTTP client for the Java plugin
      server.py                        # MCP tool definitions (49 tools)
      verification.py                  # reusable diagram/matrix verification helpers
      methodology/                     # Phase 2 pack registry + recipe runtime
        registry.py
        runtime.py
        service.py
    pyproject.toml
  plugin/                              # Java Cameo plugin
    src/com/claude/cameo/bridge/
      CameoMCPBridgePlugin.java        # Plugin entry point
      HttpBridgeServer.java            # Embedded HTTP server + routing
      handlers/
        ProjectHandler.java            # Project info, save
        ElementQueryHandler.java       # Element search, get, relationships
        ElementMutationHandler.java    # Create, modify, delete elements
        RelationshipHandler.java       # Create relationships
        MatrixHandler.java             # Native refine/derive requirement matrices
        DiagramHandler.java            # Full diagram lifecycle
        ContainmentTreeHandler.java    # Containment tree browsing
        SpecificationHandler.java      # Specification read/write
        MacroHandler.java              # Groovy script execution
      util/
        EdtDispatcher.java             # EDT dispatch with session management
        ElementSerializer.java         # Element to JSON serialization
        JsonHelper.java                # JSON parsing utilities
    plugin.xml                         # Cameo plugin descriptor
    build.gradle                       # Gradle build config
  install.sh                           # One-step installer
  LICENSE
  README.md

Development

Building the Plugin

cd plugin
./gradlew assemblePlugin -PcameoHome="/path/to/CatiaMagic"

The output goes to plugin/build/plugin-dist/com.claude.cameo.bridge/.

Running the MCP Server Standalone

python -m cameo_mcp.server

This starts the MCP server on stdio. It will fail to connect to the Java plugin unless Cameo is running with the plugin loaded.

Adding a New Tool

  1. Add the HTTP handler method in the appropriate Java handler class
  2. Register the route in HttpBridgeServer.registerHandlers()
  3. Add the async client function in client.py
  4. Add the MCP tool function with docstring in server.py

The MCP tool docstrings are critical -- they are the AI's instruction manual. Include valid values, examples, common mistakes, and cross-references to related tools.

Contributing

Issues and pull requests welcome. If you're building something similar for other MBSE tools, let's talk.

Areas where contributions would be especially valuable:

  • Additional element and relationship type support
  • Bulk operation endpoints
  • Test coverage
  • Support for Cameo Teamwork Cloud projects
  • SysML v2 profile support

License

MIT

Related: MBSE Agents

Use with mbse-agents for standards-aware modeling across aerospace, defense, automotive, and medical device domains. The agents provide practitioner-level domain knowledge (ARP4754A, DoDAF, ISO 26262, IEC 62304). This bridge provides direct tool access. Together: an AI that knows the standards at the clause level AND can modify your Cameo model.

快速设置
此服务器的安装指南

安装命令 (包未发布)

git clone https://github.com/ajhcs/cameo-mcp-bridge
手动安装: 请查看 README 获取详细的设置说明和所需的其他依赖项。

Cursor 配置 (mcp.json)

{ "mcpServers": { "ajhcs-cameo-mcp-bridge": { "command": "git", "args": [ "clone", "https://github.com/ajhcs/cameo-mcp-bridge" ] } } }