# Solutionlink + eXplore Agent Kit 0.4.0

A portable local stdio MCP, typed business planning tools, and authoring recipes learned through Solutionlink and eXplore. This is an experimental web-client adapter, not a Siemens-supported public API or a hosted service connected to someone else's account.

Start at [AGENTS.md](AGENTS.md). The kit includes source, locked dependencies, synthetic tests, setup scripts, two portable skills, schemas and operating guidance. Download provenance and SHA-256 are provided at https://explore.xcelerator.us/agent.json.

## Install

Requires Node.js 22 or newer, npm, Microsoft Edge, and your own Solutionlink authoring entitlement. Windows/Edge is the exercised platform. Other operating systems and browser channels need validation.

1. Download the versioned ZIP linked by agent.json. Compare its SHA-256 to the release manifest, then extract to a new local directory. Do not run installation directly from a remote page.
2. Inspect package.json, scripts and the source you will execute. Run `npm ci --ignore-scripts`, `npm test`, `npm run smoke`, then `npm run doctor` in the extracted directory.
3. Run `npm run login`. Complete Siemens sign-in/MFA in the dedicated Edge window, then press Enter in the terminal. It verifies authentication by a native read and closes the browser. Do not run login and MCP simultaneously because the profile has one owner.
4. For each authorized catalog/project UUID taken from your Solutionlink URL, run `npm run discover -- catalog UUID` or `npm run discover -- project UUID`. Discovery reads native identity and does not grant write scope.
5. Create config.local.json using the returned identity fields. See config.example.json for shape only: every example ID and author is synthetic. Set defaultCatalogId to your import catalog. Keep enableWrites false initially. Set SOLUTIONLINK_CONFIG to the absolute local config file path in the agent's MCP environment.
6. Register the local stdio MCP using the examples in docs/installation.md. Restart/reload that MCP entry. Call solutionlink_list_contexts and solutionlink_auth_status, then read the applicable catalog, process, scenario or business case.
7. For an authorized mutation, enableWrites must be true in your local config. Reload MCP after configuration changes. Business changes use prepare, review the saved preview, then apply. Imports overwrite an existing owned editable matching solution and have their own validation/backup/readback workflow. They are not catalog or solution creation tools.

## Working tools

16 tools: authentication status; configured context list; default catalog read; owned solution export/import; offline export inspection/import inventory; process/scenario read; business case read and native options; prepared typed TIPS/deployment/ROI changes and apply; offline dated ROI arithmetic; native avatar options; billboard content blueprint.

The MCP runs on the recipient's workstation. It uses its own profile in ~/.solutionlink-agent-kit, makes native requests inside the authenticated browser, and never exposes an auth extraction tool. Local config, profiles, input exports, plans and evidence are private workstation data. Do not upload them to the distribution site. StateDir can select a separate dedicated kit state directory; never point it at your everyday browser profile.

## Boundaries

The package does not create arbitrary catalogs or projects, edit BPMN/scenarios/information models, delete/publish records, create native 3D billboards, or save eXplore scene layouts through public tools. Recipes explain these workflows and what an additional adapter must preserve. Avatar options lists metadata, not a complete visual selection. Billboard preparation creates a local blueprint, not a native billboard.

Authenticated 3D iframe embedding remains unresolved: vendor share-host startup, framing and SSO need a supported contract. A standalone BPMN viewer can display an authorized XML snapshot, but is not live 3D eXplore. Never relay startup credentials or turn off browser protections to simulate a supported embed.

## Validation and compatibility

This release is derived from a working task-specific adapter. Offline tests exercise synthetic known-good/bad cases for import content, stale plans, partial failures, replay, relationships and ROI timing/rate readback. Fresh-package MCP smoke tests exercise tool discovery, context listing and offline ROI. Recipient live sign-in, read, and a controlled non-production write/readback remain required before production use. Native web endpoints and Angular services can change without notice. Stop when identity, ownership, schema or readback differs; do not retry an interrupted write blindly.

See docs/business-specifications.md, docs/modeling.md, docs/explore.md and docs/troubleshooting.md. Keep authored process proposals and illustrative ROI assumptions clearly labeled for domain/finance review.