Skip to main content
Glama

lettzalens

Ask questions of the Lettza dataset without hosting a copy of it. lettzalens is a small local forwarding client for the hosted Lettza MCP service: it serves the tool list from your own machine using the captured definitions, then sends each query upstream and hands the service's answer back exactly as it arrived.

Everything on the surface is read-only. You can learn the dataset's shape and where it came from, search its rows for a piece of text, pull the rows matching an exact key, place several values side by side, and summarise or rank its numeric columns. lettzalens keeps no local copy of the data and computes nothing itself — it is a lens, not a mirror.

The tools

  • dataset_columns — Every column in the dataset, which ones are numeric, the row count and the provenance banner. The right first call when you do not yet know the schema. No arguments.

  • dataset_search — Rows with a cell containing the query text, matched case-insensitively, up to 50 rows. query is required; limit optionally narrows the result (1–50).

  • dataset_row — Rows where one column equals a value exactly, compared case-insensitively. column and value are both required.

  • dataset_compare — Rows whose column matches any of the supplied values, returned in the order the values were given — the shape of an "X vs Y" question. Requires column and a values list of 2–10 entries.

  • dataset_stats — count, min, max, mean, median and sum for a numeric column. Grouping commas and currency symbols are understood; rows that do not parse as numbers are left out and counted separately. column is required.

  • dataset_top — Rows ranked by a numeric column, highest first — "which is the most X". column is required; limit caps the list (1–50) and ascending: true flips it to lowest first.

  • dataset_provenance — The dataset's source, the date it was computed, its licence and the citation to use when quoting a figure. No arguments.

Related MCP server: geolens-mcp

Install and run

As an MCP server via npm:

npx -y @reneturcios/lettzalens mcp

MCP client configuration:

{
  "mcpServers": {
    "lettzalens": {
      "command": "npx",
      "args": ["-y", "@reneturcios/lettzalens", "mcp"]
    }
  }
}

As a container:

docker run -i ghcr.io/cloudcorprecords/lettzalens
docker run --rm ghcr.io/cloudcorprecords/lettzalens dataset_provenance '{}'

CLI

A single binary covers both the command line and the MCP server. Every tool is a subcommand whose one argument is a JSON string of that tool's inputs:

lettzalens dataset_columns '{}'
lettzalens dataset_search '{"query":"lettza"}'
lettzalens dataset_row '{"column":"city","value":"Fresno"}'
lettzalens dataset_compare '{"column":"city","values":["Fresno","Modesto"]}'
lettzalens dataset_stats '{"column":"score"}'
lettzalens dataset_top '{"column":"score","limit":10,"ascending":true}'
lettzalens dataset_provenance '{}'
lettzalens --version

Exit codes: 0 for a successful forwarded call, 1 when the result that comes back is an error, and 2 for usage mistakes like an unknown subcommand or a JSON argument that does not parse.

Configuration

  • LETTZA_MCP_URL — Replaces the upstream endpoint. Defaults to the pinned hosted endpoint, https://lettza.com/mcp. (This override variable is a reconstruction introduced by this rebuild.)

The Lettza registry record lists no headers for its remote, so lettzalens never attaches an authentication header — every query goes upstream keyless.

When the upstream cannot answer — it returns a JSON-RPC error, answers with a non-2xx status, sends a body that cannot be read, or the request itself fails — the caller receives an error result whose text begins with lettzalens error:.

How forwarding works

The first query of a process opens a session against the upstream: an initialize request (protocol version 2025-03-26), a notifications/initialized notification, and then the tools/call. If the service issues an Mcp-Session-Id header during initialize, lettzalens repeats it on subsequent requests and keeps reusing that one session until the process ends. Replies are understood in both plain JSON and server-sent-events framing. Successful results are passed through untouched, isError flag and any extra fields included.

Development

npm install
npm run build   # tsc -> dist/ (dist/cli.js is the bin)
npm test        # tsx --test test/*.test.ts — stubbed upstream, no network
npm run dev     # run the CLI from source

The suite compares the served tool list against the captured surface tool for tool, walks the session handshake and its reuse, proves no Authorization header is sent under any environment, and confirms that arguments travel upstream unchanged and results return unchanged.

License

MIT — see LICENSE. Copyright (c) 2026 Rene Turcios.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      B
      maintenance
      Provides read-only SQL query access to PostgreSQL mirrors of Vizru spreadsheets on Neon, exposing REST endpoints and MCP tools for relational filtering and schema discovery.
      -
    • A
      license
      A
      quality
      B
      maintenance
      Read-only access to a self-hosted GeoLens geospatial catalog: dataset search, schemas, GeoJSON features, saved maps, and sandboxed read-only SQL over PostGIS.
      9
      302
      Apache 2.0
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables MCP-compatible clients to inspect schemas, sample rows, run safe read-only SELECT queries, plan joins, and analyze columns across SQLite and PostgreSQL with row limits, access controls, and sensitive-column masking.
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables read-only querying of the hosted Limslane dataset, including schema inspection, text and exact-value search, row comparison, column statistics, ranking, and provenance retrieval without local caching or computation. It forwards queries to the hosted Limslane MCP service and returns results verbatim.
      MIT