ArangoDB MCP Server

This is an implementation of the Model Context Protocol for ArangoDB.

Overview

To be filled.

Components

Resources

Tools

Query Tools

• readQuery

  • Execute read-only query on the database

  • Input:

    • databaseName (string): The database to query

    • aql (string): The read-only AQL query to execute

  • Returns: Query results as array of objects

• readWriteQuery

  • Execute query on the database

  • Input:

    • databaseName (string): The database to query

    • aql (string): The AQL query to execute

  • Returns: Query results as array of objects

• listDatabases

  • List all the databases on the ArangoDB server

  • Returns: Array of the databases names

• listCollections

  • List all the collections in an ArangoDB database

  • Input:

    • databaseName (string): The name of the database

  • Returns: Array of objects { "name": "<collectionName>" }

Usage

To connect to an arangodb instance running on localhost:2434, to the database "account", add the following to your claude_desktop_config.json, assuming the path to this project is /home/yourcoolname/arango-mcp-server:

Development

Clone the repository. Install everything. Setup the dev environment. Run the watcher. Edit index.ts.

Go to http://localhost:5173/ to see the inspector.

Todo

• Properly study the spec to see if the current implementation of resources actually make sense (I don't think it does)

  • The resource templates make sense

• Change all the "arango" to "arangodb" (repo name included...)

• Add back the arangodb password

• Proper README

  • Tools/resource/etc following the format of the official anthropic stuff

• Figure out notifications

• Health checks

• More tools?

• Access all the databases running on an arangodb instance

• Release on npm somehow so it can be used with npx

• resources/subscribe and notifications/resources/list_changed and resources/unsubscribe

• Properly document tools in the readme

• Like on the SQLite MCP client

  • write_query tool separated from read_query -> actually is readWriteQuery

  • list_collections (see list_tables)

• Client pool ie one client per database

• Dev environment

• resources/read with a template to read any document by database name, collection, id.

• Add username and passwords as parameters of the command