PlugWiki
obsei logo

obsei MCP Server: Setup for Claude, Cursor and Other Apps

OfficialLocal
GetPyPIobseiobsei/obseiDocs

Lets Claude, Cursor and other assistants answer questions about customer feedback stored in your own obsei setup, with citations.

obsei collects customer feedback from reviews, tickets and surveys, strips personal data when it is stored, and keeps it on your own systems. Its MCP server lets your assistant search that feedback, count it, list themes and propose issues for a person to approve.

You need an existing obsei store and its key. The server runs on your computer, and text returned to the assistant is redacted and never carries author names.

Good for

  • Product teams who collect reviews, tickets and surveys with obsei
  • Support leads who want themes and trends without exporting data
  • Privacy-minded teams who keep feedback in their own infrastructure

Works in

12 should work

Tap an app to see how to set it up and any known problems.

Install obsei MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the obsei MCP server to Claude Code. Run this command:
    
    claude mcp add obsei -e OBSEI_DB=/path/to/obsei.duckdb -e OBSEI_DB_KEY=<OBSEI_DB_KEY> -- uvx --from "obsei>=1.0" obsei mcp
    
    Ask me for OBSEI_DB, OBSEI_DB_KEY and only put the values in that config, nowhere else.
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires uv, An obsei project with a data store.

Read-only mode: Over stdio the store is opened read-only, except for the moment a proposal is saved.

Obsei is not in any tool's official plugin directory yet, so the plugin routes install from the obsei GitHub repo. The plugin routes add the server plus a voice-of-customer skill. The config routes run uvx --from "obsei>=1.0" obsei mcp, which needs uv and an existing obsei store: set OBSEI_DB to its path and OBSEI_DB_KEY to its key. The README's Cursor and VS Code buttons use the same command.

Built from the package data we hold. Not yet tested by us.

Tools

  • Counts feedback and average rating

    Counts feedback records and averages ratings, grouped by source, sentiment, intent, language, rating, route or time period.

    Try: “How many 1-star reviews did we get per week last month?”

    Files · feedback_stats

  • Finds matching feedback

    Returns matching feedback records, newest first, as citable evidence. Text has personal data redacted and never includes author names.

    Try: “Find feedback that mentions slow checkout.”

    Files · search_feedback

  • Looks up one record by ID

    Returns a single feedback record by its ID so a citation can be checked.

    Try: “Show me record rec_123 so I can verify that quote.”

    Files · get_feedback

  • Lists recurring themes

    Returns stable themes with trends, sources, languages and intents. Themes with too few people behind them are hidden for privacy.

    Try: “Which themes are rising this month?”

    Files · list_themes

  • Queues an issue for a person to approve

    Stores a proposed GitHub, Jira or Linear issue with cited record IDs in obsei's own approval queue. Nothing is sent until a person approves it with the obsei command line. Works only if such a sink is set up.

    Try: “Propose an issue about the login timeouts, citing the three worst reviews.”

    Write · propose_issue

Read from the source code

Permissions

low risk · 3 things to watch

Low risk

The assistant can read your redacted customer feedback and queue issue proposals, but nothing is sent to GitHub, Jira or Linear until a person approves it. Customer text can contain instructions, so treat answers as quotes and not commands.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • Customer text can carry hidden instructions

    Anyone can write a review. The docs mark each item as untrusted and tell agents to quote it, not follow it, but keep the assistant's other tools limited when you read feedback. github.com

  • It can queue issue proposals

    propose_issue writes to obsei's own approval queue and sends nothing out. A person approves or rejects each one with obsei approvals. github.com

  • Over stdio the store opens read-only

    It opens read-write only for the moment it stores a proposal. If obsei serve or obsei run holds the store, send proposals through obsei serve's /mcp instead. github.com

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Sign-in and keysThe store key is passed in OBSEI_DB_KEY. The optional HTTP server uses a bearer token in OBSEI_API_TOKEN.

Questions

Is the obsei MCP server official?

Yes. It is built into obsei itself and started with the obsei mcp command, from the obsei/obsei repo.

What do I need before I can use it?

An obsei project with collected feedback, the path to its DuckDB store in OBSEI_DB and the store key in OBSEI_DB_KEY. The server reads that store.

Can it send issues to GitHub or Jira by itself?

No. propose_issue only queues a proposal. A person approves it with obsei approvals approve, and only then is the issue opened.

Does it show customer names?

No. The docs say results contain redacted text, labels, source, time and record ID, and never author pseudonyms.

Is there a remote version?

Yes. obsei serve exposes the same tools at /mcp over HTTP. Set OBSEI_API_TOKEN and clients send it as a bearer token.

How do I add it to Claude Code?

Use the plugin route in the install box, which adds the server and a voice-of-customer skill from obsei's own marketplace. It is not in Claude's official plugin directory yet.

Information

updated Nov 2, 2026

Publisher
Official
Package
obsei
Registry
PyPI
Transport
stdio, Streamable HTTP
Auth
token in env
Version
1.1.1
License
Apache-2.0
Last commit
Oct 30, 2026
Open issues
4
Listed in
Official MCP registry

WebsiteDocsGitHubPyPIOfficial MCP RegistryChangelogIssues

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 3
  1. mcp.md · github.com
  2. README.md · github.com
  3. mcp_server.py · github.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.