Saagar Patel
MCP

Connect an agent to my work

Saagar Patel · public MCP servers · read-only, registry-listed, one command to connect

Some of the work in this workshop is also reachable by an AI agent directly, over the Model Context Protocol. Each server below is public, read-only, and listed in the official MCP registry under the io.github.saagpatel namespace, so you can add one to Claude, Cursor, or any MCP client with a single line.

Read-only is the whole point: none of these take a URL or a path argument, so there is nothing to point at your own machine, and the hosted ones bake their data into the deploy and make zero network calls at runtime. The two hosted servers also publish an Ed25519-signed discovery manifest, so a client can verify the bytes came from me.

Hosted servers

Streamable-HTTP endpoints you can point a client straight at, or run locally over stdio with npx if you prefer no network hop.

portfolio-mcp

The whole site as a corpus: search the essays, book, and notes, pull a page's full text, read my profile, and list the projects. It is what powers an agent reading this workshop end to end.

https://mcp.saagarpatel.dev/mcp

npx -y saagar-portfolio-mcp

Registry io.github.saagpatel/portfolio-mcp · source · npm

operant-mcp

The OPERANT benchmark as a queryable API: model calibration results, the methodology, and the case library. Ask it to compare two models, or to score your own agent against the published profiles.

https://operant.saagarpatel.dev/mcp

npx -y saagar-operant-mcp

Registry io.github.saagpatel/operant-mcp · source · npm · results & methodology

Local servers

These run on your machine over stdio. They either write files or read your local setup, so they are deliberately not hosted. Run them with uvx, no install step.

mcpforge

A scaffold generator that spins up a complete, tested FastMCP server from one plain-English sentence: the boilerplate I got tired of writing by hand, turned into one command. Generating uses your own model provider, so it is stdio-only by design.

uvx fastmcp-builder

Registry io.github.saagpatel/mcpforge · source · PyPI

MCPAudit

A local, read-only report on the MCP servers already wired into your tools: it enumerates them, risk-scores each one, and flags injection, SSRF, and shadowing surfaces. It reads your client config and reports environment-variable names only, never their values, which is exactly why it stays on your machine.

uvx --from mcp-audits mcp-audit serve

Registry io.github.saagpatel/mcp-audit · source · PyPI

mcp-trust

A public "danger grade" registry for MCP servers: ask it for a server's A-F risk letter, the capability breakdown behind that grade, and the methodology, so you can check a server before you wire it in. The grades are baked from real sandboxed scans, so it answers offline with no network calls. Registry site and PyPI release last checked from this page on 2026-07-03.

uvx mcp-trust mcp-serve

Registry io.github.saagpatel/mcp-trust · source · PyPI · registry site

shadow-mcp

A local, read-only inventory of the MCP servers actually configured on this machine: it discovers them across your clients and risk-grades each one, so you can see your real shadow-MCP surface. It reads your local setup, which is exactly why it stays on your machine.

uvx shadow-mcp mcp-serve

Registry io.github.saagpatel/shadow-mcp · source · PyPI

How to look around