Connect your agent
Add the hosted Starkscan MCP server to Claude, Cursor, Claude Desktop, Cline, or Codex.
Connect your agent
The launcher @starkscan/mcp connects any MCP client to Starkscan. It runs the bundled CLI underneath and forwards your API key from the environment. Get a key first with Get your first API key, then pick your client below. After connecting, see the MCP tools reference.
Shared environment
export STARKSCAN_CHAIN="SN_MAIN"
export STARKSCAN_API_KEY="<your key>"
# Optional: only set this for preview or self-hosted hosts.
# export STARKSCAN_BASE_URL="https://preview.example.com/api"The launcher injects https://api.starkscan.co when STARKSCAN_BASE_URL is unset and reads STARKSCAN_API_KEY from the environment — never put the key in a config file you commit. STARKSCAN_BASE_URL is the API base used for REST and launcher probes; hosted MCP appends /mcp to that base, so do not set it to a full /mcp URL or append another /api.
Generate a config
@starkscan/mcp can print a ready-to-paste config for your client, so the docs never drift from the launcher:
npx -y @starkscan/mcp print-configFor interactive setup you can use the default @starkscan/mcp; unattended configs
should pin the exact version resolved from npm after release proof.
Client recipes
All unattended recipes use command: "npx", args: ["-y", "@starkscan/mcp"], and pass STARKSCAN_API_KEY plus STARKSCAN_CHAIN. Add STARKSCAN_BASE_URL only for preview or self-hosted hosts. Before saving an unattended config, resolve npm latest, replace the package spec with that exact version, and treat the 0.4.0 verification target as available only when its registry resolution and post-publish proof agree.
Cursor
Add .cursor/mcp.json in your project:
{
"mcpServers": {
"starkscan": {
"command": "npx",
"args": ["-y", "@starkscan/mcp"],
"env": {
"STARKSCAN_CHAIN": "SN_MAIN",
"STARKSCAN_API_KEY": "${STARKSCAN_API_KEY}"
}
}
}
}Open Cursor Settings → MCP and refresh the starkscan server.
Claude Desktop
Edit claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"starkscan": {
"command": "npx",
"args": ["-y", "@starkscan/mcp"],
"env": {
"STARKSCAN_CHAIN": "SN_MAIN",
"STARKSCAN_API_KEY": "${STARKSCAN_API_KEY}"
}
}
}
}Restart Claude Desktop, then look for starkscan under the connectors menu.
Cline
Add .cline/mcp.json with the same mcpServers block as above, then reload Cline's MCP servers.
Claude Code and Codex
These have first-class CLI commands — see the one-line claude mcp add and codex mcp add recipes in the MCP quickstart.
Transports
| Transport | When | Notes |
|---|---|---|
| stdio (local) | Local clients spawning the launcher | JSON-RPC over stdio; the launcher runs the bundled binary. |
| Hosted HTTP | Remote/server clients | POST https://api.starkscan.co/mcp with X-Starkscan-Api-Key. A single JSON-RPC request returns 200 with application/json; notifications return 202; GET/DELETE return 405. |
The hosted /mcp endpoint on api.starkscan.co is authenticated with X-Starkscan-Api-Key and is origin-allowlisted: a request that sends a browser Origin must be on the allowlist or it is rejected with 403. App origins may expose /api/mcp as a compatibility route, but env-based public clients should use API-base plus /mcp. Server-to-server calls without an Origin header pass through.
Agent discovery
Agents can bootstrap from machine-readable context:
| Artifact | URL | Auth |
|---|---|---|
| Agent memory | /.well-known/starkscan-agent.json | Public |
| llms.txt / llms-full.txt | /llms.txt, /llms-full.txt | Public |
| Agent context | /v1/meta/agent-context | API key |
| Capabilities | /v1/meta/capabilities | API key |
| Chains | /v1/meta/chains | API key |
These advertise the auth header, the /v1 API-host base, the default chain, and the recommended first calls.