Installing Solana MCP in Claude Desktop, Cursor, and Cline (5 min)

MCP is the protocol, but every client has its own config file and its own quirks. This post walks through getting the techkern Solana MCP server installed in each of the six clients we officially support, with the exact JSON to paste, where to put it, and what to do when it does not show up.

Prerequisites

• Node.js 20+ (any LTS works, we ship to npm)
• A techkern API key — sign up free at techkern.xyz/signup; hobby tier is 1K queries/day
• The client of your choice installed

That is it. The server is a single npm package, fetched lazily by your client.

Claude Desktop

File: ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows.

{
  "mcpServers": {
    "techkern": {
      "command": "npx",
      "args": ["-y", "@techkern/mcp-server"],
      "env": {
        "TECHKERN_API_KEY": "tk_live_..."
      }
    }
  }
}

Restart Claude Desktop fully (quit, do not just close the window). Open a new conversation. The hammer icon in the input bar should show 10 new Solana tools (query_solana_swaps, query_pump_fun_launches, query_token_holders, query_kol_wallets, mint_authority_check, dev_token_history, whale_tracker, new_token_alerts, liquidity_depth, holder_distribution).

If tools do not show up: check Claude Desktop's developer console (View → Toggle Developer Tools) for an MCP connection error. The most common cause is a malformed JSON config — make sure your trailing commas are clean.

Cursor

File: ~/.cursor/mcp.json

{
  "mcpServers": {
    "techkern": {
      "command": "npx",
      "args": ["-y", "@techkern/mcp-server"],
      "env": { "TECHKERN_API_KEY": "tk_live_..." }
    }
  }
}

Cursor picks up MCP config on restart. To verify, open Cursor's Agent panel and the techkern tools should appear in the tool list.

Cursor-specific gotcha: if you run Cursor as a code-editing IDE and want the agent to call query_solana_swaps for in-editor research, make sure your agent's system prompt explicitly mentions that Solana MCP tools are available. Cursor's default agent prompt does not prompt the model to look at non-coding tools.

Cline (VS Code extension)

In VS Code: Cmd+Shift+P → "Cline: Open MCP Settings". This opens ~/Documents/Cline/MCP/cline_mcp_settings.json (path varies slightly per OS).

{
  "mcpServers": {
    "techkern": {
      "command": "npx",
      "args": ["-y", "@techkern/mcp-server"],
      "env": { "TECHKERN_API_KEY": "tk_live_..." },
      "disabled": false,
      "autoApprove": ["query_solana_swaps", "query_pump_fun_launches", "holder_distribution", "mint_authority_check"]
    }
  }
}

The autoApprove array lets Cline call read-only tools without per-call confirmation, which makes interactive research much faster. We recommend auto-approving every query_* and *_check / *_distribution tool — they are all read-only and idempotent.

Continue (JetBrains, VS Code)

File: ~/.continue/config.json. Add to the mcpServers array (or create it):

{
  "mcpServers": [
    {
      "name": "techkern",
      "command": "npx",
      "args": ["-y", "@techkern/mcp-server"],
      "env": { "TECHKERN_API_KEY": "tk_live_..." }
    }
  ]
}

Continue reloads MCP config on save — no restart needed.

Zed

File: ~/.config/zed/settings.json. Inside assistant:

{
  "assistant": {
    "version": "2",
    "mcp_servers": {
      "techkern": {
        "command": "npx",
        "args": ["-y", "@techkern/mcp-server"],
        "env": { "TECHKERN_API_KEY": "tk_live_..." }
      }
    }
  }
}

Zed surfaces MCP tools in its assistant panel. Same drill — the 10 techkern Solana tools appear in the tool list.

OpenAI Agents SDK (TypeScript)

If you build agents programmatically rather than using a desktop client, use the Agents SDK's MCPServerSse adapter:

import { Agent } from "@openai/agents";
import { MCPServerSse } from "@openai/agents/mcp";

const techkern = new MCPServerSse({
  name: "techkern",
  url: "https://api.techkern.xyz/mcp",
  headers: { Authorization: `Bearer ${process.env.TECHKERN_API_KEY}` },
});

const agent = new Agent({
  name: "Solana Research Bot",
  model: "gpt-4.1",
  mcpServers: [techkern],
  instructions: "You research Solana using techkern MCP tools — DEX swaps, pump.fun launches, holder distributions, KOL wallets.",
});

const result = await agent.run("Find pump.fun launches with >100 holders in the last hour.");
console.log(result.output);

The SSE transport means you do not spawn a local subprocess — the agent connects directly to our hosted MCP endpoint. Lower latency, less local config.

Verifying the install

Once installed in any client, the smoke test prompt is:

> "What MCP tools do you have available from techkern?"

The model should enumerate the 10 tools. If it cannot, the server is not connected — check the client's MCP log.

After that:

> "Use query_pump_fun_launches to find the three most-funded launches in the last 10 minutes."

You should get structured results back in under 250ms. If you do, your install works end-to-end.

Rate limits to know

• Hobby (free): 1,000 queries/day, no card
• Pro: 50,000 queries/day, $19/mo or pay-per-query in $TECH
• Scale: unlimited, custom

Rate limit is per API key. Hitting it returns 429 with Retry-After — your MCP client will surface that as a tool error, which the model can read and adapt to (e.g. "we have hit our daily quota, switching to a different signal").