MCP server

pip install "sysml2kit[mcp]" and run:

sysml2kit mcp serve                    # stdio (default)
sysml2kit mcp serve --transport http   # streamable HTTP

Client config:

{"mcpServers": {"sysml2kit": {"command": "sysml2kit", "args": ["mcp", "serve"]}}}

Every tool takes model file paths (.json interchange always works; .sysml needs the parse extra), returns a JSON dict with a status key, and reports failures as {"error": ..., "status": "failed"} instead of raising. Artifacts are returned as file paths, not payloads.

Tool	What it does
`model_show(path, traceability)`	Element tree, kind counts, optional trace matrix
`model_validate(path)`	S2K rule issues with severity counts
`model_diff(path_a, path_b, by_name)`	Element-level differences (capped list)
`model_export(path, out, to, stable_ids)`	Convert to interchange JSON or `.sysml`
`model_diagram(path, out, kind)`	Write a mermaid `.mmd` (trace or tree view)
`requirements_trace(path)`	Matrix plus unsatisfied/unverified lists
`requirements_extract(path)`	`RequirementSpec` list (the adapter payload)
`library_load(name, out)`	Write a packaged rf-library model as JSON

A typical agent loop: write a .sysml file, model_validate it, fix issues, requirements_trace to check coverage, model_export --stable-ids to commit the interchange form, model_diagram to explain the result.