Portfolio MCP Server

Exposing this site's data as MCP tools any AI client can query.

Cloudflare Workers · MCP SDK · TypeScript · Zod

Results

10
MCP tools covering experience, projects, hiring info, quality metrics, and writing: 2
transports — remote (Streamable HTTP/SSE) and local (stdio): 5 min
edge cache — new site data propagates with no server redeploy

Problem

Recruiters and AI assistants can only learn about a candidate by scraping HTML — brittle, lossy, and disconnected from whatever the candidate actually maintains. I wanted any MCP-aware AI client to be able to query my experience, projects, and skills as accurate, structured data instead.

Approach

Built a Cloudflare Worker using the agents SDK's McpAgent, exposing 10 tools (experience, projects, skills, background, contact, hiring logistics, quality metrics, mentoring, and the blog) over both Streamable HTTP and legacy SSE.
Added a prebuild step to this site that exports the same data.ts/posts.ts modules it renders from as static JSON — single source of truth, no duplicated resume data between the site and the server.
Validated every fetched payload with Zod schemas, since the data crosses a network boundary between two separately deployed repos.
Shipped both a remote transport (zero-install, via Claude Code/Desktop) and a bundled stdio entry point for local use.

How it flows

Site build

lib/data.ts

JSON export

public/data/*.json

Cloudflare Worker

fetch + 5min cache

MCP client

Claude, Cursor…

From the code

src/tools.ts

MCP SDK

export function registerPortfolioTools(server: McpServer) {
  server.registerTool(
    "get_experience",
    {
      title: "Get work experience",
      description:
        "Ashish Kumar's work experience history — company, role, dates, responsibilities, quantified results, and business impact for each position.",
    },
    async () => jsonResult(await getExperience())
  );
  // … 6 more tools
}

Real excerpt · view source →

Outcome

A deployed MCP server that any MCP client can query for structured information about my background — verified end-to-end with the MCP Inspector CLI against both transports before calling it done.

Business Impact

A working demonstration of the exact skill I'm selling — deliberate AI-tooling integration — not just a bullet point claiming I know MCP.