MCP servers¶
Two Model Context Protocol servers put Flexo in front of an AI assistant — one for each of Flexo's APIs. Both are Python, built on FastMCP with the streamable HTTP transport, and both default to read-only so an assistant can explore a model without being able to change it until you say so.
Layer 1 MCP server¶
Wraps the Layer 1 REST API as MCP tools.
- Repository: Open-MBEE/flexo-mms-layer1-mcp
| Tools for | Covering |
|---|---|
| Structure | Reading and managing organizations, repositories and branches |
| Models | Reading, loading, querying and committing RDF with SPARQL |
| Version control | Locks, and diffs between commits |
| Access control | Policies, groups and collections |
With READ_ONLY=true (the default) only the read and query tools are registered.
MMS_URL names the Layer 1 endpoint. The Authorization header on an MCP request is
forwarded to Layer 1, so the assistant acts as the user who is logged in, with that user's
policies applied.
docker run -d -p 8000:8000 \
-e MMS_URL=https://your-flexo-mms-server \
-e READ_ONLY=false \
flexo-mms-layer1-mcp
SysML v2 MCP server¶
Wraps the SysML v2 API as MCP tools.
- Repository: Open-MBEE/flexo-mms-sysmlv2-mcp
SYSMLV2_URL names the API; READ_ONLY=true (the default) exposes only the GET
endpoints; MCPPATH moves the MCP endpoint from /mcp. Bearer tokens from the MCP client
are forwarded to the API.
Then point an MCP client at http://localhost:8000/mcp, with an Authorization: Bearer …
header if the deployment needs one.