Connect Koji to Cursor: MCP Setup Guide for Product Engineers

Cursor's Model Context Protocol (MCP) support turns the editor into a research-aware coding environment. By connecting Koji's MCP server to Cursor, your AI assistant gains direct access to live customer interview transcripts, structured answers, themes, quality scores, and full study reports — without copy-pasting a single quote.

This guide walks through the full setup, the most useful Koji + Cursor workflows, and the gotchas to know about authentication and tool selection.

Why Connect Koji to Cursor

Most teams already use Cursor to write code. Most teams also have customer feedback scattered across Slack threads, Notion docs, and a forgotten interview tool. The MCP integration closes that gap: Cursor can query your Koji workspace the same way it queries the filesystem.

Concretely, this unlocks four high-value workflows for product engineers:

• Context-aware feature work. Before adding a new onboarding step, ask Cursor "what did the last 20 interview participants say about onboarding friction?" and it pulls the actual quotes.
• PRDs grounded in evidence. Generate a one-page spec that cites real interview transcripts instead of speculative bullet points.
• Triage with proof. When grooming the backlog, Cursor can rank tickets against the themes appearing most frequently in published Koji reports.
• Post-launch sense-check. After shipping, ask Cursor to summarise interviews collected since the deploy and flag regressions in sentiment or quality scores.

This is a wedge no traditional research tool can offer. SurveyMonkey, Typeform, and Qualtrics still expect you to log into a separate dashboard. Koji's MCP server brings the data to where engineers already work.

Prerequisites

Before you start, make sure you have:

• A Koji workspace with at least one published study and a few completed interviews to query.
• Cursor 0.45 or later. Earlier versions do not support MCP.
• A Koji API key. Generate one in Settings → API Keys in your workspace. See Managing API Keys for guidance on scopes and rotation.
• Familiarity with editing JSON config files. The setup is one short paste.

The Koji MCP server is included on every paid plan and on the Free plan for evaluation. Tool calls are not metered separately — they share the standard interview credit pool described in Plan Comparison Guide.

Step 1: Open Your Cursor MCP Config

Cursor stores MCP server definitions in a JSON file. The location depends on your operating system:

• macOS / Linux: ~/.cursor/mcp.json
• Windows: %USERPROFILE%\.cursor\mcp.json

If the file does not exist, create it. You can also open the same config from inside Cursor via Cursor → Settings → MCP → Edit Config.

Step 2: Add the Koji MCP Server Block

Paste the following into the mcpServers object of your mcp.json. Replace YOUR_KOJI_API_KEY with the key you generated above.

{
  "mcpServers": {
    "koji": {
      "url": "https://koji.so/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_KOJI_API_KEY"
      }
    }
  }
}

If you already have other MCP servers configured (GitHub, Linear, filesystem, etc.), add the koji entry alongside them rather than replacing the file.

Save the file and restart Cursor. The first time you open a chat after the restart, Cursor will list koji in its connected servers panel and surface its 15 tools.

Step 3: Verify the Connection

Open a new chat in Cursor (Cmd/Ctrl + L) and type:

@koji list my recent studies

Cursor should call koji_list_studies and return your study titles, statuses, and interview counts. If it reports an authentication error, double-check the bearer token in mcp.json — a stray space or quote is the most common culprit.

For a deeper smoke test, try:

@koji summarise the most common themes from the most recent published study

This chains koji_list_studies, koji_get_study, and koji_get_interviews automatically. The model picks the right tool for each step — you do not need to name them explicitly. See MCP Tool Reference for the full list.

The 15 Koji Tools You Now Have in Cursor

Once connected, Cursor can call any tool the Koji MCP server exposes:

Read tools — koji_list_studies, koji_get_study, koji_get_interviews, koji_get_transcript, koji_get_account, koji_get_study_data, koji_get_report

Write and analysis tools — koji_create_study, koji_update_brief, koji_publish_study, koji_generate_report, koji_publish_report, koji_configure_study, koji_export_data, koji_import_respondents

The read tools are safe to call freely. The write tools modify your workspace, so configure Cursor's tool-approval policy to require confirmation before they run. See MCP Authentication and Security.

Recommended Cursor + Koji Workflows

The integration works best when you anchor it to a concrete coding task. Here are five patterns teams ship with on day one.

1. Writing a feature spec from interview evidence

Open the spec doc and prompt Cursor:

Draft a one-page PRD for a new onboarding checklist. Cite at least
five direct quotes from the most recent onboarding study in Koji,
and group user pain points by theme.

Cursor will fetch transcripts and structured answers, then generate the spec with the quotes inline. The structured answers come from Koji's six question types (open_ended, scale, single_choice, multiple_choice, ranking, yes_no) — see Structured Questions in AI Interviews for how those become chartable data the model can reason over.

2. Composer-driven ticket grooming

Open Composer (Cmd/Ctrl + I) on your tickets/ directory and prompt:

For each open ticket in this folder, score 1–5 how strongly the
last 30 days of Koji interviews support shipping it. Cite the
themes that match.

The model walks each ticket, calls koji_get_study_data for theme aggregations, and writes the score plus a short rationale into the ticket file.

3. Pre-deploy regression check

Before merging, ask Cursor to scan recent interview themes for any spike in negative sentiment. Useful when you ship onboarding, pricing, or auth changes — exactly the surfaces participants comment on most.

4. Generating a stakeholder summary

After a study concludes, prompt:

Generate a stakeholder report for the latest study, then publish it.

This chains koji_generate_report and koji_publish_report. The published report URL comes back inline in Cursor — paste into Slack and you are done. This is the same flow described in Real-Time Research Insights, but driven from your editor.

5. Creating a quick discovery study from a code TODO

Highlight a // TODO: validate this with users comment, open chat, and prompt:

Create a 5-minute Koji study to validate this assumption with 20
mobile users. Use one open_ended question and one scale question
for confidence.

Cursor calls koji_create_study and returns a shareable interview URL. From idea to live study in under a minute.

Troubleshooting

"Tool not found" errors. Restart Cursor — MCP servers are only registered on startup. If the issue persists, run cat ~/.cursor/mcp.json | jq . to confirm the JSON is valid.

Authentication failures. Bearer tokens are sensitive; never paste them into a shared .cursor config. Use Cursor's environment-variable substitution ("Authorization": "Bearer ${env:KOJI_API_KEY}") and load the key from your shell.

Tool selection drift. If Cursor calls koji_get_interviews when you wanted koji_get_study_data, name the tool explicitly in your prompt. The model improves with corrections inside the same chat.

Rate limits. The Koji MCP endpoint shares the standard API rate limits documented in Rate Limits and CORS. Heavy bulk reads should batch by limit and cursor.

How This Compares to Pasting Transcripts

Some teams still copy interview snippets into Cursor by hand. That works for a single decision but breaks down at scale: stale data, missing quality scores, and no way to filter by theme or sentiment.

The MCP integration eliminates all three. Cursor sees the same authoritative data your researchers see, scoped by your API key, with quality scores that automatically suppress low-effort responses (only interviews scoring 3 or above are surfaced — see How the Quality Gate Works).

That is the difference between "AI with vibes" and "AI with evidence."

Related Resources

• Koji MCP Integration Overview
• Connect Koji to Claude (Setup Guide)
• MCP Tool Reference
• MCP Authentication and Security
• MCP Best Practices
• MCP Workflow Guide for Product Managers
• Structured Questions in AI Interviews