Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@DiffBeam MCP Server validate a 4F beam expander with 50mm and 200mm lenses"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
DiffBeam is a deterministic Gaussian beam optics service for AI agents and engineering tools. Its anonymous MCP searches optical component catalogs and reports server/catalog revisions. A separate credentialed Direct Agent Bridge analyzes and applies revisioned edits to a user's Gaussian Beam planner workspace.
DiffBeam does not host or proxy a language model. Bring any compatible AI client.
Endpoint: https://catalog-api.gaussian-beam.com/mcp/
Release: 2.0.0
Registry name: com.gaussian-beam/diffbeam
Human-readable documentation: https://gaussian-beam.com/mcp/
Machine-readable manifest: https://gaussian-beam.com/.well-known/mcp/server.json
Contracts: https://gaussian-beam.com/mcp/contracts/
MCP clients connect directly with Streamable HTTP. The public endpoint is anonymous, stateless, and read-only. It has no solver, design-link, or workspace authority.
Related MCP server: zemax-mcp-server
An internet-capable AI that can send HTTPS POST requests does not need native MCP installation. The planner's Connect your AI dialog creates a workspace-specific endpoint, temporary scoped credential, and ready-to-copy prompt. The normal flow is:
get_workspace
make one small planner-native operation such as add_lens or set_wavelength
inspect the returned beam feedback and explain the visible revision
refine with the next useful operation; use dry_run only for explicit preview-only work
Read the Direct Agent Bridge guide. Never put a workspace credential in a URL, public page, model configuration, log, or repository.
get_diffbeam_state
search_fibers
search_fiber_collimators
search_lenses
The bridge is not part of the anonymous MCP. It uses the temporary capability from Connect your AI and groups operations into:
workspace reads, changes, and revision history;
deterministic solving, beam summaries, waist finding, and point measurement;
fiber, collimator, and lens catalog searches;
planner-native source, element, parameter, unit, and beam-target mutations;
undo and revision restore.
Writes require the current expected_revision and create a live planner
revision by default. The public MCP cannot access these operations.
The JSON file in examples/ is a planner-import example for documentation and
bridge workflows. It is not an argument accepted by the anonymous catalog MCP.
This public repository contains discovery metadata and integration documentation. It deliberately contains no production credentials, private workspace data, telemetry, or deployment configuration. The documents and examples are licensed under Apache-2.0.
Security issues should be reported privately as described in SECURITY.md.
This server cannot be deployed
AI-callable calculators and engineering models with real formulas. No hallucinated math.
Parametric 3D CAD for AI agents: build print-ready parts, check them, export STL, 3MF or STEP.
Deterministic reasoning stack for AI agents: simulate, decide & compute, plus cross-domain tools.
Design, save, and run outcome-aligned AI workflows and verifiers, with reliable image output.
