Reference

MCP servers behind this workspace

The skills in this library are one half of how an agent works the Siemens toolchain; the other half is the MCP servers they call. Every server here is a private GitHub repo under chrisdamonmartini, same account as this skills library, verified by reading each repo's own git remote get-url origin and counting tool declarations directly in source — not from memory.

Inventory

Server (`.mcp.json` name)	Repo	Tools	What it does
`tc`	`tc-mcp`	15	Instance-agnostic Teamcenter bridge over the JSON REST SOA surface — one server, any TC web tier via named profiles (`tc_connect(profile=...)`). Ships a bundled 1,592-operation SOA catalog for offline lookup (`tc_search_operations`, `tc_service_operations`) so a route can be found without a live connection.
`nx`	`nx-mcp`	19	Drives NX 2606 headless via NXOpen Python — part creation, expressions, journals, export, screenshots, a persistent `run_journal` worker, and a full API index (25,792 classes) for offline lookup.
`capital`	`capital-mcp`	11	AI↔Capital System Architect bridge. `materialize_platform` turns an AI-authored design bundle into an importable Capital project XML with every component placed and every pathway wired — no manual canvas work. Also drives the Signal Hub and Connectivity Hub plugins.
`made`	`made-mcp`	10	MADe (reliability/RAMS) REST client bridge — async job pattern for FMEA, RBD, and failure-model work.
`amesim`	`amesim-mcp`	17	Drives Simcenter Amesim headless via `ame_apy`, Amesim's own bundled Python (version-locked — the bridge self-configures `sys.path`/DLL dirs at launch rather than running under a normal interpreter). Builds and runs 1D multiphysics models, reads results.
`cameo`	`CameoTcConfig`	19	The Cameo↔Teamcenter connector configuration tool as an MCP layer — switch environments/releases, diagnose the connector, and (via `cameocfg/sniffer.py`) record the live Cameo↔TC wire traffic for reverse-engineering undocumented calls.
`cameo-model`	`cameo-mcp` (`server.py`)	31	Headless Cameo/MagicDraw model authoring — Jython run inside a real Cameo instance via a `CommandLineAction` plugin, backed by a persistent worker (0.4s/call vs. ~40s cold start). Elements, diagrams, stereotypes, TC push/pull, recording/replay.
`cameo-ops`	`cameo-mcp` (`ops_server.py`)	10	The operational half of the same repo — check-in/out, environment switch, model scaffolding, teardown. Split from `cameo-model` because these are project/session lifecycle actions, not model-authoring calls.
`roadmap`	`roadmap-mcp`	14	Client for `roadmap.xcelerator.us`, the cross-project work tracker. List/add/update items, per-lane status, weekly self-reported status, GitHub Actions dispatch for connected sites.
`nx-docs`	`nx-docs-mcp`	3	Search/read the NX knowledge-base site's content directly (list topics, search, get one).
`sc3d-docs`	`sc3d-docs-mcp`	3	Same pattern for the Simcenter 3D knowledge base.
`amesim-docs`	`amesim-docs-mcp`	3	Same pattern for the Amesim knowledge base.
`capital-docs`	`capital-docs-mcp`	3	Same pattern for the Capital knowledge base.
`made-docs`	`made-docs-mcp`	3	Same pattern for the MADe knowledge base.

Thirteen repos serve fifteen registered server names (cameo-model/cameo-ops share one repo, two entry points).

How the registry works

xcelerator-workspace/.mcp.json.template is the single source of truth for which servers exist and how to launch them — every entry is a relative path under the workspace root, so the same template works on any machine that has the sibling repos checked out next to it. scripts/Install-XceleratorWorkspace.ps1 materializes the real .mcp.json from the template. Edit the template, never the generated file — the same discipline as skills being edited at their canonical source and republished, not hand-edited at the destination.

Getting these onto a remote agent

Unlike the skills library, none of these are published as an installable Claude Code plugin — they're wired in purely through .mcp.json, and there is no read-only web fallback equivalent to this Developer section for browsing tool schemas without a connection. A remote agent needs:

GitHub access to clone xcelerator-workspace plus whichever sibling server repos it needs (all private, same account).
Local resources some servers depend on and GitHub can't provide: amesim needs an actual Simcenter Amesim install at a known path (its Python interpreter is the Amesim-bundled one, version-locked); nx needs a running NX session; cameo/cameo-model/cameo-ops need a real Cameo/MagicDraw install with the companion plugin loaded.
**Credentials from %USERPROFILE%\.xcelerator\credentials\** (DPAPI-encrypted, Windows-only, tied to the creating account+machine) — these do not travel with the repos and have to be created fresh on the target machine. See tc-mcp/profiles.json's credFile convention for the pattern every server that needs Teamcenter credentials follows.

Related skills

tc-soa-session (the session/auth pattern the tc server implements), tc-capture-awc-calls (the technique behind cameo's wire-recording feature), cameo-headless-automation (the plugin/worker pattern behind cameo-model).

Generated from references/mcp-servers.md in the tc-automation-skills library, which is the canonical copy and also serves as the agent skill set for Teamcenter work.