designsafe-mcp
Agent-first cyberinfrastructure layer for DesignSafe that enables reproducible computational science workflows (geotechnical earthquake engineering with OpenSees), from planning and validation through approved execution, with provenance and tested snippet grounding.
README
designsafe-mcp
An agent-first cyberinfrastructure layer for DesignSafe, scoped to a
bounded subset of the quoFEM pipeline for geotechnical earthquake
engineering with OpenSees as the forward solver. The server exposes
scientific actions, find a tested workflow, build and validate a job, price
it, run it after human approval, and package the result with provenance,
so an AI agent (or a researcher) composes reproducible computational science
instead of driving low-level APIs. supported_capabilities() declares
exactly what is supported, in which scientific domain, and how much of
it is tested; requests outside that map are declined by design.
dapi and Tapis remain the substrate; this layer translates research intent into trustworthy workflows. Orchestration comes from a tested snippet corpus (executed, self-checking, version-pinned examples), never from a model's recall of the API.
ARCHITECTURE.md records the full design: the layer stack, why decisions
live in tools rather than in the model, deployment topologies, and the
evaluation methodology.
Tools
| Action | Tool |
|---|---|
| Decide how to run it | plan_simulation(request, facts...) walks the OpenSees decision matrix |
| See the matrix itself | opensees_matrix(); the deck's slide ships as an MCP resource |
| Find a tested workflow | search_snippets(query) |
| Search the corpus | search_community(query) over UW community data, dapi examples, the ds-workflows book |
| Ground in the real app interface | describe_app(app_id) |
| Move data | stage_inputs(local_dir) |
| Construct the experiment | build_job_request(...), build_workflow_preview(...) |
| Check before spending | validate_job(job), estimate_cost(job) |
| Human gate | approve_submission(job) -> token; submit_job(job, token) refuses without it |
| Execute and monitor | submit_job, job_status, get_results |
| Provenance | write_manifest(...) |
Run
uv venv .venv && uv pip install -p .venv/bin/python -e .
.venv/bin/python -m designsafe_mcp.server # stdio MCP server
Register it once for Claude Code with
claude mcp add designsafe -- $PWD/.venv/bin/python -m designsafe_mcp.server,
or add the same command to .jupyter/mcp_settings.json for jupyter-ai.
Auth rides on dapi's environment; nothing is stored here.
demo/transcript.txt holds an executed transcript: discovery, grounding,
staging, validation, cost, a refused unapproved submission, the approved run,
results, the manifest, and a compiled two-stage DAG preview.
Evals
.venv/bin/python evals/runner.py --mode planner # deterministic floor
.venv/bin/python evals/runner.py --mode agent --models haiku,sonnet --trials 3
Golden cases in evals/cases.yaml map natural-language requests to the
expected decision; agent mode drives real models against the server in
DESIGNSAFE_MCP_MOCK=1 mode (no Tapis calls, no SUs) and scores each
trace on decision, grounding, and the approval gate. Pass rate per case
per model is the ability metric.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.