Core question: can a small adapter expose a GraphQL endpoint as a clean MCP tool surface without turning into a giant integration platform?
Desired demo:
- define one GraphQL endpoint;
- expose 1–2 operations as MCP tools;
- call the tools from a simple local setup;
- show how adapter logic maps inputs/outputs.
Scope:
- one provider / one schema sample;
- minimal auth handling;
- clear request/response mapping;
- local demo only.
Non-goals:
- generic GraphQL platform;
- enterprise auth matrix;
- hosted control plane;
- broad schema introspection framework.
Core question: can a narrow input format generate a useful workflow stub faster than hand-writing boilerplate?
Desired demo:
- input a small structured spec;
- output a workflow scaffold with stages/placeholders/comments;
- show where human editing still matters.
Scope:
- one tiny JSON/YAML schema;
- documentation-first experiment or tiny CLI;
- 2-3 sample input/output pairs;
- explicit notes on limits.
Non-goals:
- autonomous workflow design;
- full orchestration engine;
- provider-specific control plane;
- vague "AI builds everything" claims.
Core question: can a lightweight utility inspect an MCP server and summarize what it exposes in a human-friendly way?
Desired demo:
- point tool at a server;
- list tools/resources/prompts;
- generate a compact markdown summary.
Only experiments that are:
- understandable quickly,
- relevant to AI devtools / MCP,
- demoable,
- and not better as a standalone repo from day one.