Most databases do one thing. Postgres is relational. MongoDB is document. Neo4j is graph. Redis is key-value. ArcadeDB does all of them — graph, document, key-value, time-series, and vector search — in a single engine. It is the spiritual successor to OrientDB, the multi-model database that SAP acquired in 2017. Written in Java, embeddable, and capable of running SQL, Cypher, Gremlin, and GraphQL queries against the same data. Now it has two MCP servers in the registry.
What ArcadeDB Is
ArcadeDB's core proposition is consolidation. Instead of running Postgres for relational data, Neo4j for graph traversals, MongoDB for documents, and a vector database for embeddings, you run one engine that speaks all four models natively. The data is stored once and queryable through whichever model makes sense for the access pattern. A social network can model relationships as a graph, store user profiles as documents, cache session data as key-value pairs, and run vector similarity searches over embeddings — all in a single ArcadeDB instance.
The project has 737 stars, 95 forks, and 15 watchers. Apache-2.0 license. 52 active commit weeks in the past year, 10 contributors, 15 releases per year. The maintenance profile is as active as projects with ten times the star count. It supports SQL (including full-text search), Apache TinkerPop Gremlin (graph traversal), OpenCypher (Neo4j's query language), GraphQL, and MongoDB-compatible queries. It can import from OrientDB, Neo4j, and relational databases.
The MCP Integration
For AI agents, the combination is particularly powerful. An agent connected to ArcadeDB can traverse a knowledge graph to understand entity relationships, query documents for unstructured data, and run vector similarity searches for semantic matching — all through a single MCP server. This collapses what would normally require three or four separate tool calls into different databases into one tool call against one engine.
The practical use case: an AI agent building a research workflow can store documents, build a knowledge graph of entities mentioned in those documents, and run semantic searches over embeddings, all without managing multiple database connections or data synchronization. The multi-model approach eliminates the impedance mismatch between specialized databases.
Two Entries, Two Scores
| Entry | Namespace | Score | Has Package |
|---|---|---|---|
| com.arcadedb/mcp-server | com.arcadedb | 77 | Yes |
| io.github.ArcadeData/arcadedb | io.github.ArcadeData | 69 | No |
The same underlying project, two registry entries, an 8-point gap. The difference is instructive.
Why the Gap
The com.arcadedb entry scores 77 because it has an installable package, which the provenance category rewards. The io.github.ArcadeData entry scores 69 because it relies on repo-based installation only. Both share the same source repo, the same maintenance signals, the same popularity metrics. The 8-point spread comes almost entirely from the package presence and namespace differences.
This is a pattern we see repeatedly in the registry: the same project with multiple entries scoring differently based on metadata completeness. The scoring model does not penalize having multiple entries, but it does measure each one independently. For project maintainers, the lesson is clear — if you publish a package, register the entry that references it.
The Score Breakdown (com.arcadedb)
| Category | Weight | Assessment |
|---|---|---|
| Provenance (30%) | Good | Source repo, Apache-2.0 license, unique description, installable package. Namespace does not match GitHub owner. |
| Maintenance (25%) | Perfect | 52 active commit weeks, 10 contributors, 15 releases/year |
| Popularity (20%) | Decent | 737 stars, 95 forks, 15 watchers |
| Permissions (25%) | Clean | No secrets, no flags, no credential concerns |
| Overall | 77 |
No flags on either entry. No secrets required. Apache-2.0 license. The one provenance detail worth noting: the com.arcadedb namespace does not match the GitHub organization (ArcadeData), which slightly reduces the namespace-matching signal. This is common for projects that use a domain-based namespace rather than a GitHub-based one — it is not a red flag, just a weaker provenance signal than a direct match.
Context
Database servers have been part of the MCP registry since early days, but most are connectors to mainstream databases — Postgres adapters, MySQL bridges, MongoDB wrappers. ArcadeDB is different because the multi-model approach means one MCP server replaces what would otherwise require multiple database-specific servers. For agent architectures that need structured data, graph traversal, and vector search, a single multi-model tool is more efficient than orchestrating calls across three specialized ones.
The OrientDB lineage matters too. OrientDB was one of the early multi-model databases, acquired by SAP and eventually deprecated. ArcadeDB carries that architectural vision forward with a modern codebase and active development. 52 active commit weeks is the maximum possible — this project ships code every single week.
Score: 77 (com.arcadedb) / 69 (io.github.ArcadeData). No flags. Apache-2.0.
Sources: ArcadeData — GitHub · Scorecard: com.arcadedb (score 77) · Scorecard: io.github.ArcadeData (score 69)