AI Levels


Disclaimer: table generated with [Grok](https://grok.com/), except by the last column.

| AI IDE / Tool | Company | Free Tier / Version                          | Pricing (Paid) |Pros | Cons | My notes |
|--------------------|---------------------------|----------------------------------------------|------------------------------------|-----------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------| -- |
| [Cursor](https://cursor.com/)                     | [Anysphere](https://anysphere.inc/)         | Hobby: 2,000 completions/mo + 50 slow premium requests; 14-day Pro trial | $20/user/mo (Pro), $40/user/mo (Business) | Best-in-class agentic features (Composer/Agent mode), excellent multi-file edits & repo understanding, fast UX, supports multiple models (Claude, GPT, Gemini, etc.) | Free tier very limited after trial, higher price than some rivals, can feel unpredictable on very complex refactors | Using it for work |
| [Claude Code](https://claude.com/product/claude-code) | [Anthropic](https://www.anthropic.com/)     | Limited free tier / API credits; terminal-focused | ~$17–20/mo (Pro equivalent via API usage or plans), pay-as-you-go | Exceptional reasoning/debugging, massive 200K+ context, ideal for complex logic/architecture, integrates well into VS Code/terminal | More CLI/agent than full visual IDE, costs scale with heavy usage, no native multi-file editor UI | Planning to use it for work. Everyone in my bubble is migrating to claude |
| [Google Antigravity](https://antigravity.google/) | [Google](https://antigravity.google/) | Yes — full public preview (generous limits, no hard cutoff yet); some Pro features emerging | Free (preview); ~$20/mo Pro expected soon | Completely free access right now (Gemini 3 Pro + others), huge context (1M+ tokens), strong multi-agent planning, browser/automation skills | Preview-stage bugs & quota complaints (e.g., refresh inconsistencies), Google ecosystem lock-in, maturity still catching up | Using for personal projects |
| [GitHub Copilot](https://github.com/features/copilot) | [GitHub / Microsoft](https://github.com/features/copilot) | Limited free (students/open-source); 50 requests/mo trial in some cases | $10/mo (Individual), $19–39/mo (Business/Enterprise) | Mature & reliable, works in many IDEs (VS Code, JetBrains, etc.), excellent GitHub integration, model choice (Claude, GPT, Gemini) | Less fully agentic than dedicated IDE forks (more autocomplete + chat), subscription needed for full power | Second option when no credits on antigravity. On free tier, it is dumber than antigravity |
| [Windsurf](https://windsurf.com/) | [Windsurf](https://windsurf.com/) (originally Codeium) | Yes — generous for individuals: 25 credits/mo, unlimited basic completions, limited Cascade/agent flows | $15/mo (Pro: 500 credits), $30/user/mo (Teams) | Strong value & speed, great for prototyping & budget users, solid multi-file & autocomplete, often called best free-capable option | Credit limits hit heavy users quickly on free tier, slightly behind Cursor in some deep reasoning/agent benchmarks | Planning to use it for personal projects | 

Also planning to play with vscode + [qwen](https://github.com/QwenLM) + ([cline](https://marketplace.visualstudio.com/items?itemName=saoudrizwan.cline) or [continue](https://marketplace.visualstudio.com/items?itemName=Continue.continue))

AI IDEs


- [LLM Arena](https://beta.lmarena.ai/leaderboard/text)
- [artificialanalysis.ai](https://artificialanalysis.ai/)
- [Agent Arena](https://gorilla.cs.berkeley.edu/leaderboard.html)

Hope you find the best one for your case!


AI Leaderboards


The idea here is not explain what MCP is, just show a sample:

# Dependencies

Install it with npm/yarn/pnpm:

- @modelcontextprotocol/sdk
- zod

# Code

```javascript
import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import {
  CallToolRequestSchema,
  ListToolsRequestSchema,
} from "@modelcontextprotocol/sdk/types.js";
import { z } from "zod";

const tools = [
  {
    name: "Sum",
    description: "do a sum",
    inputSchema: z.object({ x: z.number(), y: z.number() }),
    handler: (args) => ({
      content: [
        { type: "text", text: `${args.x} + ${args.y} = ${args.x + args.y}` },
      ],
    }),
  },
];

const server = new Server(
  { name: "mcp-server", version: "0.0.1" },
  { capabilities: { tools: {} } }
);
server.setRequestHandler(ListToolsRequestSchema, async () => {
  return {
    tools: tools.map((t) => ({
      name: t.name,
      description: t.description,
      parameters: t.inputSchema,
    })),
  };
});

server.setRequestHandler(CallToolRequestSchema, async (request) => {
  const { name, arguments: args } = request.params;

  try {
    const tool = tools.find((t) => t.name === name);
    if (!tool) {
      throw new Error(`Tool unknown: ${name}`);
    }
    return await tool.handler(args);
  } catch (error) {
    console.error(`Error executing tool ${name}:`, error);
    throw error;
  }
});

const main = async () => {
  const transport = new StdioServerTransport();
  await server.connect(transport);
};

main();
```

# Sample calling

List tools:

```sh
echo '{"jsonrpc": "2.0", "id": 1, "method": "tools/list"}' | pnpm dlx tsx cli.ts
```

Executing operation:

```sh
echo '{"jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": {"name": "Sum", "arguments": {"x": 1, "y": 2}}}' | pnpm dlx tsx cli.ts
```

For http server check it [here](/blog/posts/2025-04-28-mcp-http);


A basic MCP sample for CLI with typescript

MCP sample - CLI


The idea here is not explain what MCP is, just show a sample:

# Dependencies

Install it with npm/yarn/pnpm:

- @modelcontextprotocol/sdk
- zod

# Code

```javascript
import express, { Request, Response } from "express";
import { SSEServerTransport } from "@modelcontextprotocol/sdk/server/sse.js";
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { z } from "zod";

///////////////////////////////////////////////////////////////////////////////
const tools = [
  {
    name: "Sum",
    description: "do a sum",
    inputSchema: z.object({ x: z.number(), y: z.number() }),
    handler: (args: { x: number; y: number }) => ({
      content: [
        {
          type: "text" as const,
          text: `${args.x} + ${args.y} = ${args.x + args.y}`,
        },
      ],
    }),
  },
];

const mcpServer = new McpServer({ name: "mcp-server", version: "0.0.1" });

for (const tool of tools) {
  mcpServer.tool(
    tool.name,
    tool.description,
    tool.inputSchema.shape,
    tool.handler
  );
}

///////////////////////////////////////////////////////////////////////////////
const app = express();

// to support multiple simultaneous connections we have a lookup object from
// sessionId to transport
const transports: { [sessionId: string]: SSEServerTransport } = {};

app.get("/sse", async (req: Request, res: Response) => {
  const transport: SSEServerTransport = new SSEServerTransport(
    "/messages",
    res
  );
  const sessionId = transport.sessionId;

  transports[transport.sessionId] = transport;
  res.on("close", () => {
    delete transports[sessionId];
  });
  await mcpServer.connect(transport);
});

app.post("/messages", async (req: Request, res: Response) => {
  const sessionId = req.query.sessionId as string;
  const transport = transports[sessionId];
  if (transport) {
    await transport.handlePostMessage(req, res);
  } else {
    res.status(400).send("No transport found for sessionId");
  }
});

const port = process.env.PORT || 3000;
app.listen(port, () => {
  console.log(`HTTP Server running on port`, { port });
});
```

# Test

Run server:

```sh
pnpm dlx esno ./sample.ts
```

Connect to SSE to create a session and receive responses (keep it running):

```sh
curl -N http://localhost:3000/sse
```

To list tools:

```sh
curl --request POST \
  --url 'http://localhost:3000/messages?sessionId={SESSION ID}' \
  --header 'content-type: application/json' \
  --data '{"jsonrpc": "2.0", "id": 1, "method": "tools/list", "params": {}}'
```

To do a call:

```sh
curl --request POST \
  --url 'http://localhost:3000/messages?sessionId={SESSION ID}' \
  --header 'content-type: application/json' \
  --data '{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "Sum",
    "arguments": {
      "x": 1,
      "y": 2
    }
  }
}'
```

For a cli sample check it [here](/blog/posts/2025-04-28-mcp-cli);


A basic MCP http server sample with typescript

MCP sample - HTTP


- Chroma (Clickhouse)
- Elastic Search
- Faiss
- Milvus
- Mongo
- postgres / pgvector
- Qdrant
- Redis / redisearch
- Rektor
- Vespa
- Weaviate

Then I started to compare them on [grok](https://grok.com/chat/daa86a5b-ec2d-439a-a4ff-76a4d5d73b38)


AI Levels

AI IDEs

AI Leaderboards

MCP sample - CLI

MCP sample - HTTP

Vector DBs