Skip to content

MCP server

The Bastyn MCP server lets an AI agent look up two things for an AI agent you are planning: which regulations and standards may apply, and which technology categories it will need. Both tools are read-only and free. Nothing you send is stored and no AI model is called.

MCP server URL (Streamable HTTP)
https://mcp.bastyn.ai/mcp

Anyone can connect and list the tools without signing in. Calling a tool needs a signed-in Bastyn user: your client opens a browser sign-in the first time you call one.

Let your agent add it

Paste this into your agent (Claude Code, or any agent that can add MCP servers). It adds the server, lists the tools and stops.

Add the Bastyn MCP server to yourself and check it works.
URL: https://mcp.bastyn.ai/mcp (Streamable HTTP, OAuth sign-in).
If you are Claude Code, run: claude mcp add --transport http bastyn https://mcp.bastyn.ai/mcp
Then list the tools the server offers and report their names and titles. Do not call any tool yet.
If a sign-in is needed, tell me and wait.

A good answer lists exactly two tools: check_regulations_and_standards and recommend_tech_stack.

Add it yourself

  • Claude.ai: Settings > Connectors > Add custom connector, paste the URL, choose Sign in when needed (the auto-detected “No sign-in” is wrong for tool calls), press Connect, then enable it in the chat.
  • ChatGPT: Settings > Plugins > Add > Add custom MCP server, paste the URL, pick OAuth or no authentication.
  • Claude Code: run claude mcp add --transport http bastyn <URL above>, then /mcp to sign in.
  • MCP Inspector: run npx @modelcontextprotocol/inspector@latest --server-url <URL above> --transport http. The server has no dynamic client registration, so turn on Client Settings > OAuth Client ID Metadata Document and give it the HTTPS URL of a metadata file that lists the Inspector’s callback (http://localhost:<port>/oauth/callback).

Try the tools

No field is required, but tell the agent something about your AI agent first. Anything you leave out is reported as a gap (“tell us X”), never as “does not apply”.

Check regulations and standards. A good answer gives each regulation a status (applies, likely applies, can’t tell, does not apply) with a reason, and each matched standard a fit rating: sector, region or general. “Plus N more…” means the summary was cut and the full list is in the structured result.

Use the Bastyn "Check regulations and standards for an AI agent" tool.
My agent is a chatbot that answers retail customers' questions about crypto assets and suggests portfolio ideas. Domain: finance. It is used in the EU and UK. Users are customers. It handles personal and financial data. It is an assistant. Do not guess anything I did not say. Show me the result and list what the tool says it could not tell.

Recommend technology categories. A good answer marks each category applicable, not needed or can’t tell. Applicable categories list candidate products, with featured ones first. They are catalogue entries, not recommendations, and “no data” for vulnerabilities does not mean low risk.

Use the Bastyn "Recommend technology categories for an AI agent" tool.
My agent is a chatbot that answers retail customers' questions about crypto assets and suggests portfolio ideas. Domain: finance. Users are customers. It uses retrieval (rag) and tools. Model provider: anthropic. Do not guess anything I did not say. Show me the categories and anything the tool says it could not tell.

If something goes wrong

  • 401 on a tool call: you are not signed in. Sign in (in Claude Code run /mcp; in Claude.ai press Connect and check it says Sign in when needed).
  • 502 right after a deploy: wait a minute and try again.
  • “Incompatible auth server” in the Inspector: turn on its Client ID Metadata Document setting.

Results are a rules-based indication from the answers you give, not legal advice.

For what the connector receives and keeps, see the MCP connector privacy page.