Skip to content

Latest commit

 

History

History
61 lines (46 loc) · 1.7 KB

File metadata and controls

61 lines (46 loc) · 1.7 KB

EXPERIMENTS PLAN — launchery-lab

Experiment 1 — MCP → GraphQL bridge

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.

Experiment 2 — Workflow auto-generation

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.

Experiment 3 — MCP discovery tool

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.

Selection rule

Only experiments that are:

  1. understandable quickly,
  2. relevant to AI devtools / MCP,
  3. demoable,
  4. and not better as a standalone repo from day one.