MCP server

Survey Coder Pro runs a hosted Model Context Protocol server. Connect it to Claude.ai, Cursor, or any MCP-aware client and code surveys conversationally — no glue code required.

Endpoint

https://api.surveycoder.io/mcp

Authentication is the same API key you use elsewhere (x-api-key header). The server speaks SSE-over-HTTP and is reachable from any MCP transport that supports remote servers.

Tools available

Tool	What it does
code_responses	One-call coding — verbatims in, codes out. Mirrors POST /v1/code.
list_projects	List projects in your org.
get_question	Read a question’s codebook and coded responses.
generate_codebook	Generate a codebook from a sample of responses without coding everything.
import_codebook	Push a codebook into an existing question.
estimate_cost	Free estimate of how many credits a job will cost.
wait_for_job	Long-poll an async job until it finishes.
read_usage	Current credit balance and recent usage.

The MCP server is read-mostly by default — destructive operations (delete project, revoke key) are not exposed. Use the REST API or dashboard for those.

Connect from Claude.ai

1. Open Claude.ai and go to Settings → Connectors.
2. Click Add custom connector (or Add remote MCP server).
3. Name it Survey Coder Pro and paste the URL https://api.surveycoder.io/mcp.
4. When prompted, paste your API key — Claude will send it as x-api-key on every tool call.
5. Start a new chat and ask: “Code these 8 verbatims about laundry detergent…”

Connect from a custom MCP client

TypeScript

import { Client } from '@modelcontextprotocol/sdk/client/index.js';
import { SSEClientTransport } from '@modelcontextprotocol/sdk/client/sse.js';

const transport = new SSEClientTransport(
  new URL('https://api.surveycoder.io/mcp'),
  { requestInit: { headers: { 'x-api-key': process.env.SCP_API_KEY! } } }
);

const client = new Client(
  { name: 'my-app', version: '1.0.0' },
  { capabilities: {} }
);

await client.connect(transport);

const tools = await client.listTools();
const result = await client.callTool({
  name: 'code_responses',
  arguments: {
    question: 'Which brand of laundry detergent do you prefer and why?',
    responses: ['Tide because it removes stains better', '...'],
  },
});

Python

from mcp import ClientSession
from mcp.client.sse import sse_client
import os

async with sse_client(
    "https://api.surveycoder.io/mcp",
    headers={"x-api-key": os.environ["SCP_API_KEY"]},
) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()
        tools = await session.list_tools()
        result = await session.call_tool(
            "code_responses",
            arguments={
                "question": "Which brand of laundry detergent do you prefer and why?",
                "responses": ["Tide because it removes stains better", "..."],
            },
        )

Conversational example

A typical chat with the MCP server connected:

You: Code these 8 verbatims about toothpaste preference: [pastes list]

Claude: I’ll use code_responses… (calls the tool). Here’s the codebook — 5 codes across 2 categories. The dominant theme is whitening (4 of 8), followed by mint flavor (3 of 8). One verbatim mentioned sensitivity. Want me to drill into any category?

The model can chain calls — for example, estimate_cost before code_responses, or wait_for_job after kicking off a large batch.

Errors

The MCP server returns the same structured errors as the REST API. The full error code and doc_url come through in the tool response so the model can correct itself or escalate to you.

Note

Rate limits and credits work identically to the REST API — every coded response is one credit, regardless of whether it came through MCP or HTTP.

Related

API reference

Every MCP tool maps to a REST endpoint. See the spec.

Coding pipeline

What happens under the hood. Read more.