What it does
This MCP resource connects the DAG Studio causal-graph canvas to an AI agent through browser-native WebMCP tools. The agent can read the live graph, including nodes, roles, positions, directed edges, exposure and outcome, and add or remove nodes and edges. Its changes appear on the same canvas the researcher edits, trigger a visible notification and enter the same undo history. Cycles, duplicate edges and invalid node references are rejected. The existing causal engine reports open backdoor paths, minimal adjustment sets and identifiability, and a separate tool checks whether a proposed adjustment set blocks those paths without including descendants of the exposure. Other tools generate R (dagitty) or Python (NetworkX) code for the graph and run a linear Gaussian simulation with summaries and a preview. Graph, analysis and undo history stay client-side.
A researcher would reach for it when working on a causal DAG and wanting an agent to inspect the graph, test a candidate confounder or mediator, or compare structural implications of competing assumptions, while the researcher keeps control of what is accepted. The publisher frames every agent edit as a proposal for review, not as a finding.
Running it requires a browser with WebMCP support, such as ChatGPT's WebMCP-capable browser, or Chrome with the WebMCP testing flag enabled. The header shows a connected status when the tools are available. The publisher hosts a static HTTPS sandbox and releases the code under the Apache 2.0 licence. It was built for the OpenAI 2026 WebMCP Challenge on top of an existing DAG Studio application.
Limitations and requirements
Requires a WebMCP-capable browser (ChatGPT's browser or Chrome with the enable-webmcp-testing flag). Tools are available only when the header shows WebMCP connected. The publisher states that causal structure is an assertion, not a fact discovered by software, and that results depend on the encoded assumptions; the engine and agent do not determine scientific truth.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
DAG Studio WebMCP
Read and edit a live DAG Studio canvas through its predefined WebMCP tools.
v1.0.0As shown on registry.modelcontextprotocol.io when we last read it