Back to Directory/Developer Tools

io.github.Axius-SDC/sdcmodeler

The modeler's client for SDCStudio: search the catalog, estimate, assemble and publish SDC4 models.

Developer ToolsPythonv4.3.1

sdcmodeler

The modeler's client for SDCStudio. Search the public catalog, estimate, assemble and publish an SDC4 data model, generate its package and an application. A Python client with httpx as its only dependency, a CLI, and sdcmodeler mcp, an MCP server for any agent.

License Python

Why

An agent that builds a feature decides a data model at that moment. With sdcmodeler the data is modeled in SDC by default: the agent searches the catalog first, reuses published components wherever one fits, mints only what is new, publishes the model, and takes the package. The result is interoperable because it is composed from shared, published components, and verifiable because every record validates against a published schema.

The name follows the role pattern of the other Semantic Data Charter servers: sdcvalidator validates, sdcgovernance governs, sdcreader reads, sdcmodeler models. It also names what SDCStudio calls every user: a Modeler.

Install

pip install sdcmodeler

The catalog and estimates need no account. Building needs an SDCStudio API key, created in Settings, API Keys, with the scopes and spending limit the holder should have:

export SDCSTUDIO_API_KEY=sdcs_...

The three layers

The client

from sdcmodeler import SDCStudioClient

c = SDCStudioClient()                       # key from $SDCSTUDIO_API_KEY, URL from $SDCSTUDIO_URL
hits = c.search_components("blood pressure", type="xdquantity")
tree = {
    "label": "Vitals",
    "components": [
        {"ct_id": hits["results"][0]["ct_id"]},            # reuse: free
        {"label": "Pulse", "data_type": "XdCount"},        # mint: 100 credits
    ],
    "clusters": [],
}
estimate = c.estimate(tree, "Vitals")       # the server's estimate: reuse, mint, problems, total_credits, wallet, key, estimate_id
result = c.assemble("Vitals", tree, estimate_id=estimate["estimate_id"])   # 500 credits plus the mints

Every leaf existing: the server assembles, publishes and generates the package in one call (result["mode"] == "reuse"; published and generated say so). Some leaves new: the pipeline runs and result["job_ct_id"] is polled with c.job_status(...), or c.wait_for_job(...), until completed, when the job has published and generated as asked and its result carries the model's links. publish=False leaves a draft for a person to review. The server refuses a build whose tree changed since the estimate.

A 402 is InsufficientCredits with the numbers in credits. A 403 from the key's policy (no spend scope, or past its limit) is KeyPolicyRefused. The client never funds a wallet; a person does that at the wallet page.

The CLI

sdcmodeler search "blood pressure" --type xdquantity
sdcmodeler models "patient"
sdcmodeler model <ct_id>
sdcmodeler artifact <ct_id> xsd --dir ./schemas
sdcmodeler estimate tree.json --title "Vitals"
sdcmodeler build tree.json --title "Vitals" --wait
sdcmodeler job <job_ct_id>
sdcmodeler package <ct_id> --dir ./packages
sdcmodeler app <ct_id> --profile lightweight --dir ./apps
sdcmodeler wallet

The MCP server

sdcmodeler mcp

A JSON-RPC 2.0 stdio server, no SDK, the way sdcvalidator and sdcgovernance are built. A client configuration needs only:

{"command": "uvx", "args": ["sdcmodeler", "mcp"], "env": {"SDCSTUDIO_API_KEY": "sdcs_..."}}
ToolDoesSpends
search_componentsthe public catalog by label, type and projectno
list_models, get_modelpublished models and their artifact routesno
estimate_modelwhat a tree would reuse, mint and cost; returns estimate_idno
build_modelassemble, publish and generate (or publish: false for a draft); needs the estimate_id the person saw; refuses past the key's limityes
job_statusone job by id: status, step, cost so far, the model's linksno
download_packagea model's package ZIP with the key, including a model in a private project; builds a missing package first (free, once) unless generate is falseno
generate_appgenerate and download the application; needs confirm_charge; 1,500 credits the first time, free afteryes
walletbalance in credits and the key's policyno

The policy lives in the server. No spend without an estimate shown: build_model takes the estimate_id that estimate_model returned for the same title and tree in this process. Stop at the key's limit: a build whose estimate exceeds what the key may still spend is refused before the server is asked. Never pay: a 402 is reported with the wallet page's address and the tool stops. No tool takes a URL, a key or a payment method as an argument; the key is the environment the server starts with.

The package or the application

get_model names the artifacts: the XSD 1.1 schema, the instance template, JSON, JSON-LD, SHACL, HTML, GraphQL. The package is a contract with no runtime attached; a feature inside an existing application implements on it, in any language, validating every record locally with sdcvalidator. The generated application (lightweight or enterprise profile) is a Django implementation on the same package, right when the feature is a new service. CordovaOS is ten of them sharing data with zero integration code.

What is not here

Validation (sdcvalidator's server), governance (sdcgovernance's), receipt verification (sdcreceipt's) and bringing existing data across (sdc-agents) are their own registered servers. sdcmodeler composes with them; it does not re-implement them.

Release

Two steps: merge to main, then push the tag v<version> matching pyproject.toml. release.yml runs the tests, builds, publishes to PyPI by trusted publisher and to the MCP registry.

License

Apache License 2.0. Copyright Axius SDC, Inc.

Installation

Source-derived launch command. Check the maintainer’s required arguments and credentials before running:

bash
uvx sdcmodeler

Set up in your AI client

Merge this template into ~/Library/Application Support/Claude/claude_desktop_config.json. Keep existing servers. Add any arguments, credentials, and permissions required by the maintainer; this template has not been install-tested.

json
{
  "mcpServers": {
    "io-github-axius-sdc-sdcmodeler": {
      "command": "uvx",
      "args": [
        "sdcmodeler"
      ]
    }
  }
}

Restart Claude Desktop completely for changes to take effect. Confirm the server appears connected in the client’s tool list, then try a read-only example from its documentation.

Claude Desktop setup reference

Package

sdcmodelerpypi

Compatible MCP Clients

io.github.Axius-SDC/sdcmodeler works with any MCP-compatible client. Copy the config snippet from the Configuration section above and add it to the file shown for your client, then restart the application.

  • Claude Desktop~/Library/Application Support/Claude/claude_desktop_config.jsonRestart Claude Desktop completely for changes to take effect.
  • Cursor~/.cursor/mcp.jsonRestart Cursor for changes to take effect.
  • VS Code.vscode/mcp.jsonReload VS Code window for changes to take effect.
  • Windsurf~/.codeium/windsurf/mcp_config.jsonRestart Windsurf for changes to take effect.
  • Claude Code.mcp.jsonSave at the project root, then start Claude Code in that project and review the MCP server approval prompt. Keep real credentials out of shared files.

Learn More