Skip to main content
Glama
MongrelDB-Viewer

MongrelDB-Viewer

MongrelDB Viewer is free and open source. Use it as a local vector database GUI or expose the open database through its local vector database MCP server.

What it does

MongrelDB Viewer has two connection modes:

Capability

Direct folder

mongreldb-server

Open style

Embedded, exclusive filesystem lock

Multi-client HTTP

Browse schema and rows

Yes

Yes

Run SQL, DDL, and DML

Yes

Yes, when the server permits it

Show foreign-key graph

Yes

Not exposed by the current server adapter

Native retrieve_text provenance

Yes

No, SQL ANN fallback

Install, backfill, or rebuild ANN from the UI

Yes

No

Run REINDEX

Yes

Sent as server SQL

Agent and in-app MCP

Yes

Yes

Catalog credentials and encryption passphrase

Yes

Server auth only

Direct mode uses mongreldb-core and mongreldb-query in the application process. Server mode uses mongreldb-client against a running daemon. See connection modes before opening production data.

Related MCP server: mmnt-mcp-server

Screenshots use the bundled six-table demo database.

Deck

Schema map

Deck overview

Schema constellation

Table browser

SQL workbench

Table browser

SQL workbench

Vector search

Agent chat

Dense ANN lab

Agent chat

MCP bridge

About

MCP bridge

About

Install

Download a package, when attached, from GitHub Releases. If no package matches your platform, build from source:

git clone https://github.com/visorcraft/MongrelDB-Viewer.git
cd MongrelDB-Viewer
npm ci
npm run tauri build

Source builds need Rust 1.95+, Node.js 22+, and the native libraries required by Tauri 2. Bundles are written below src-tauri/target/release/bundle/.

See installation and builds for platform prerequisites, development mode, Linux display behavior, upgrades, and model downloads.

Five-minute tour

npm ci
npm run tauri dev

Then:

  1. Leave Direct folder selected.

  2. Choose an empty directory.

  3. Click Create demo DB.

  4. Open Deck, Stars, Table, and SQL.

  5. In ANN, search hybrid retrieval across indexes.

Demo creation writes six tables, foreign keys, representative secondary indexes, a 384-dimensional embedding column, and dense HNSW ANN. The first run may download all-MiniLM-L6-v2. If the download is unavailable, demo creation can fall back to zero vectors; use Re-embed from text column after the model becomes available.

The demo button hides after its first successful use. Reopen the demo through Recent or Direct folder on later launches. The complete walkthrough and schema are in first launch and demo.

Product surfaces

Surface

Purpose

Deck

Connection metadata, table and row totals, index capabilities, live insights, and schema-derived SQL recipes

Stars

Pan and zoom graph of database, tables, columns, indexes, and Direct-mode foreign keys

Table

Column types and flags, embedding source, index options, row samples, and REINDEX

SQL

MongrelDB/DataFusion workbench with recipes, in-memory history, result grid, and CSV copy

ANN

Direct-mode ANN install/rebuild/backfill and table-scoped semantic search

Agent

OpenAI Chat Completions-compatible tool loop over the current database

MCP

Local HTTP JSON-RPC server or separate stdio process with the same tool surface

About

Product metadata, complete license texts, dependency credits, and runtime notices

The Viewer recognizes all six public MongrelDB secondary-index families: Bitmap, LearnedRange/PGM, FM, ANN, Sparse, and MinHash.

Safety before use

  • SQL, Agent, and MCP tools are not read-only. They can run DDL, DML, REINDEX, and ANN installation without a per-statement confirmation dialog.

  • Direct mode owns an exclusive database lock. Disconnect before another exclusive process opens the root, and disconnect before filesystem backups.

  • The in-app MCP endpoint has no authentication. The UI binds it to 127.0.0.1; any local process that can reach the port can invoke its tools.

  • Clicking Save in Agent stores the base URL and model in WebView localStorage. The API key remains in process memory and is cleared on exit.

  • Prompts and tool results sent to a remote chat endpoint can contain database data.

Read the security policy and operations guide for the full trust model and local-data inventory.

Documentation

Start here:

Use the application:

Operate and extend it:

Architecture at a glance

React + TypeScript
       |
       | Tauri invoke
       v
Rust command layer
       |
       +-- Direct: mongreldb-core + mongreldb-query
       |
       +-- Server: mongreldb-client + HTTP
       |
       +-- Embeddings: fastembed MiniLM
       |
       +-- Shared tools: Agent chat + MCP HTTP/stdio

The active connection and embedding hub are shared by SQL, Agent, and in-app MCP. Stdio MCP is a separate process and must open its own Direct or Server connection. See architecture for lifetimes, data flows, trust boundaries, source layout, and extension points.

Development gate

npm ci
npm run build
cargo fmt --manifest-path src-tauri/Cargo.toml --check
cargo clippy --manifest-path src-tauri/Cargo.toml --all-targets --all-features -- -D warnings
cargo test --manifest-path src-tauri/Cargo.toml

Dependency changes also require:

scripts/regen-credits.sh

Full workflow: CONTRIBUTING.md.

Support and policy

  • Mongrel — Commercial multi-system workbench with native MongrelDB support.

  • MongrelDB Viewer — Free, open-source MongrelDB GUI and MCP server.

License

MongrelDB Viewer is available under MIT OR Apache-2.0. See LICENSE-MIT and LICENSE-APACHE.

Available Tools

9 tools
constellationC

Return the schema constellation graph (tables, columns, indexes).

Inputs: no named parameters.

database_overviewA

High-level overview of the open database: path or server URL, connection mode, tables, ANN readiness.

Inputs: no named parameters.

describe_tableA

Describe a table: columns, types, flags, and secondary indexes (Bitmap/PGM/FM/ANN/Sparse/MinHash).

Inputs: table (required).

execute_sqlC

Run a MongrelDB SQL statement (DataFusion). Prefer read-only SELECT for exploration.

Inputs: sql (required), max_rows.

install_dense_annA

Install ANN on a table (direct/local only). Default algorithm=hnsw, quantization=dense (full f32 cosine). Supported pairs: hnsw×{dense,binary_sign,product}, diskann×dense, ivf×dense. product requires product_num_subvectors (must divide dimension). Set rebuild=true to drop and recreate.

Inputs: table (required), rebuild, algorithm, dimension, +12 more.

list_embedding_providersB

List embedding providers available in the viewer.

Inputs: no named parameters.

list_tablesA

List all tables in the open MongrelDB database with row counts and index capabilities.

Inputs: no named parameters.

reindexA

Run engine REINDEX maintenance (analyze + compact + GC). Omit table for the whole database; pass table for one table.

Inputs: table.

Tool Schema Changelog

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

Loading tool schema changelog…

    TDQS

    B3.4/5.0

    Scored across 9 tools

    Disambiguation3/5

    Most tools are distinct, but schema-inspection duties overlap: list_tables vs database_overview both report tables, and describe_table vs constellation both expose columns and indexes. Descriptions hint at differences (connection-level vs graph-level), yet an agent can plausibly pick the wrong one for a simple 'what tables exist' or 'what indexes are here' question. Execution-oriented tools (execute_sql, semantic_search, install_dense_ann, reindex) are clearly separated.

    Naming Consistency3/5

    Eight of nine names are snake_case, but the verb_noun pattern is broken by noun-only entries (constellation, reindex) and a noun_noun entry (database_overview). Naming is still readable and predictable enough to infer intent, but it is not a uniform convention.

    Tool Count5/5

    Nine tools is well within the ideal 3-15 range and each maps to a real operation in a database viewer (inspect, query, index, maintain, discover providers). No tool feels redundant enough to remove wholesale, and none of the surface is padded.

    Completeness4/5

    Exploration, ANN setup, maintenance, and provider discovery are all covered, with execute_sql acting as a general escape hatch for anything missing. Minor gaps remain: no way to drop/uninstall an ANN index schema-side, no explain/query-plan tool, and no explicit listing of installed ANN indexes per column beyond describe_table.

    Maintenance

    ActivityNo data
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers