Get This Tool
Setoku
Pricing
- Model
- Free
Summary
The AI returns 92,118 fans; the right answer is 71,204 — because nobody told it which accounts to exclude. Setoku is a self-hosted MCP knowledge server built to close that gap: it forces the AI to read what your data means before it touches a single row.
The server provides read-only query access to your data alongside a persistent, human-curated layer of metric definitions and known gotchas — so when the AI asks for merchandise revenue and the data is incomplete, it flags the gap rather than returning a wrong total. Proposed changes to that knowledge layer require a person to approve them in the admin console, so a bad session cannot silently rewrite your definitions. Published dashboards run on live data at a static link, with no frontend to maintain. The ceiling appears when your data questions require joins or transformations the analytics engine cannot express, at which point you are writing custom integrations via the connect skill.
Bottom line: Pick this for a team that keeps correcting the same AI query mistakes because nobody encoded the business rules; skip it if your reporting layer already enforces metric definitions centrally and you need something the whole org can self-serve without a self-hosted box to run.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Community Benchmarks Community
Sign in to submit a benchmarkNo community benchmarks yet. Be the first to share a real-world data point.
Pros
Sign in to edit- Metric definitions and gotchas are stored and retrieved before any query runs, so the AI stops returning totals that ignore the exclusions your analysts already know about — without anyone having to re-explain the rules each session.
- Knowledge updates require explicit human approval in the admin console, so a misbehaving or injected session cannot silently corrupt the definitions the whole team relies on.
- Read-only access is enforced at the database engine level with row caps and statement timeouts, which means a runaway query cannot lock your production database or pull unbounded data.
- Published apps stay live on your data at a static link with no frontend to maintain, so a dashboard built in one session keeps working for the whole team without anyone running a deploy.
- Apache-2.0 open source with self-hosting on your own VPS, which means teams with data residency or audit requirements can inspect every layer and keep credentials entirely off external infrastructure.
Cons
Sign in to edit- There is no hosted option. Before a single query runs, your team needs a VPS provisioned, the server deployed, data sources connected, and tokens distributed. Teams without internal infrastructure ownership hit this wall immediately and move to a hosted analytics tool instead.
- The knowledge layer only improves when someone runs /setoku:curate and approves pending corrections. Teams that skip curation get a knowledge base that stagnates — the AI repeats the same mistakes on new questions because no one encoded the new definitions, which recreates the exact problem Setoku was installed to solve.
- The analytics engine is a read-only mirror of your database plus ingested lake data. Queries that require transformations or joins not expressible in that engine require a custom integration via /setoku:connect — at which point someone is writing and maintaining integration code, and the 'just ask in plain language' promise applies only to what the mirror already contains.
- App publishing is scoped to what Claude Code can generate from your data. Teams that need interactivity, custom filtering logic, or branded UI beyond what the publish_app tool produces are maintaining a separate frontend anyway, which eliminates the no-deploy advantage for anything past a basic table or chart.
Community Reviews
Sign in to write a reviewNo reviews yet. Be the first to share your experience.
About
- Platforms
- Self-hosted on Linux VPS
- API Available
- Yes
- Self-Hosted
- Yes
- Last Updated
- 2026-07-26T08:16:10.297Z
Best For
Who it's for
- Teams using Claude or other MCP clients with internal structured data
- Self-hosted environments requiring audit logs and read-only access
- Business intelligence where metric gotchas must be remembered
What it does well
- Query company metrics with automatic context for definitions and exclusions
- Publish live dashboards from natural language requests
- Reconcile data across multiple systems for consistent reporting
- Flag incomplete data sources instead of returning misleading totals
Integrations
Discussion Community
Sign in to commentNo discussion yet. Sign in to start the conversation.
Compare Setoku
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeCommunity Notes & Tips Community
Sign in to contributeBe the first to contribute. General notes, observations, gotchas, and tips from people who use this tool day-to-day.
Frequently Asked Questions
- Is Setoku free?
- Yes — Setoku is fully free to use. There is no paid tier.
- Is Setoku open source?
- Yes. Setoku is open source.
- Does Setoku have an API?
- Yes. Setoku exposes a developer API. See the official documentation at https://setoku.com for details.
- Can I self-host Setoku?
- Yes. Setoku supports self-hosting on your own infrastructure.
- What platforms does Setoku support?
- Setoku is available on: Self-hosted on Linux VPS.
Hours Saved & ROI Stories Community
Sign in to contributeBe the first to contribute. Concrete time/cost savings, with context. e.g. "Cut my code review backlog from 4h to 45m per week."
Curated lists that include this category
Setoku runs as a self-hosted MCP server that gives any MCP-compatible AI client — Claude, Codex, or similar — read-only access to your structured data plus a queryable knowledge layer describing what that data means. The core loop: the AI calls find_context or get_metric before running a query, reads the canonical definition and any recorded gotchas, then executes against a read-only analytics mirror. Row caps, statement timeouts, and an append-only audit log are enforced at the database engine level, not by parsing SQL after the fact. Claude Code skills handle setup: /setoku:onboard wires a repo, /setoku:connect adds a data source, /setoku:generate drafts business context from existing code, and /setoku:curate queues knowledge changes for a human to accept.
The differentiating feature is the correction loop. When the AI encounters a definition it doesn’t have, it can propose an update — but that proposal sits in the admin page until a person approves it, outside any agent execution path. Nothing an injected or runaway session does can silently alter the knowledge base. The vendor explicitly notes this design choice as a guard against prompt injection rewriting the system’s understanding of your metrics.
Publishing works through two MCP tools: publish_app and update_app turn a query result or dashboard into a live-data web app at a shareable link, with no deployment step and no frontend code to maintain. The demo shows this with a sponsorship pricing table and an attendance forecast, both wired to live demo data. The model does no inference on the server, so adding Setoku does not increase inference costs — it connects to whatever AI you are already running.
The self-hosted requirement is the practical constraint most teams hit first. There is no hosted option described anywhere in the documentation or vendor page; you run this on your own VPS or internal infrastructure. Each user gets a revocable token embedded in their connector URL. The Apache-2.0 license means the code is auditable and forkable, which matters for teams with strict data residency requirements — but it also means support is community-driven, not contractual.