My open MCP server for AI agents
I build MCP servers for a living, so this site has one. Point your AI assistant at it to read my projects, ask about services, and, when you ask it to, send me a brief.
Short answer
The server is at https://sakthivignesh.com/mcp (Streamable HTTP, no sign-in). It has four read-only tools (get_profile, list_projects, get_project, ask) and one write tool (submit_inquiry) that emails me a brief. It is rate limited, and I read every message myself.
Connect in one line
The address is the same for every client. There is no account, key or sign-in.
https://sakthivignesh.com/mcp
# Claude Code
claude mcp add --transport http sakthi https://sakthivignesh.com/mcp
# Cursor, Windsurf and most JSON-configured clients
{ "mcpServers": { "sakthi": { "url": "https://sakthivignesh.com/mcp" } } }
# VS Code (.vscode/mcp.json)
{ "servers": { "sakthi": { "type": "http", "url": "https://sakthivignesh.com/mcp" } } }
# Claude.ai and Claude Desktop: Settings, Connectors, Add custom connector, paste the addressWhat your agent can do
| Tool | What it does | Changes anything? |
|---|---|---|
| get_profile | Who I am, where I am based, my services and every way to contact me | No |
| list_projects | Projects I have built, filterable by category | No |
| get_project | Full details of one project: summary, stack, my role, links | No |
| ask | Answers a plain-English question about services, pricing or process, from this site's own content | No |
| submit_inquiry | Sends me a project brief on behalf of a real person | Yes: emails me |
Sending me a brief
submit_inquiry is the only tool that does anything. It needs the person's real name and email, a message of at least 20 characters, and an explicit confirmation that the person asked the agent to contact me and agreed to share those details. A good agent asks you first.
I read every brief myself and reply to the email address given. That address also gets one short confirmation (at most one per day) that says an AI assistant sent the message on their behalf and that nothing else will follow if they did not ask for it.
- Optional fields: project type, company, phone, timeline, budget and which assistant is sending it
- An identical brief sent twice is recognised and counted once
- Messages with more than three links are refused
Rate limits
The server is open, so it is limited. Read tools are generous; the one write tool is strict.
| Limit | Value |
|---|---|
| Any request, per client | 60 per minute |
| Briefs, per client | 3 per hour and 10 per day |
| Briefs, per email address | 3 per day |
| Briefs, all clients together | 40 per day |
Why it exists
A portfolio is something an AI assistant should be able to read directly, instead of scraping it. It is also the honest way to show what I build: this server is a small, real example of one with least-privilege tools, argument validation, rate limits and a human in the loop for the one action that matters.
Frequently asked questions
Is the MCP server free to use?
Yes. It is open to any MCP client, with no sign-in. It is rate limited to keep it fast and to stop abuse.
Can an AI agent hire me through it?
An agent can send me a brief on a person's behalf using submit_inquiry. I read it and reply to that person by email. The agent cannot book, pay or commit to anything.
What data does the server keep?
Read tools keep nothing. A submitted brief is stored with the details the agent supplied and the sender's network address, the same as a brief from the contact form, so I can reply and stop abuse.
Does the ask tool use a language model?
No. It answers from the site's own content and says plainly when it does not know, and then points to submit_inquiry or the contact details.
Can I build an MCP server like this for my business?
Yes. That is one of my services: MCP servers with the least access that does the job, validated inputs, rate limits and approvals on anything risky.
Keep reading
Tell me what eats your week.
My intake agent asks a few quick questions and puts your brief in front of me. I read every one.