Feature : MCP server

[ExtReMLapin]

A MCP server for LLM might have the advantage to automate testings, especially where we're trying to replicate features.

1. LLM Write queries based on documentation (Ex : Neo4J documentation)
2. Runs the same query on both ArcadeDB and original implementation (Neo4J using Neo4j MCP server) and compare the two results.

Right now on the tests I'm running, ( #3428) We're taking the bet the LLM did read correctly the documentation, but sometime its fails and just claim feature doesn't work while it works correctly, rule of thumb should be "Just behave like Neo4J" .

---

Low priority in my opinion.

[lvca]

This is a great idea, with an MCP server you could run analytics with Claude against your database.

[lvca]

It's working on branch mcp-server. From Studio you can configure it:

[lvca]

The authorization key is generated at login time to be able to be copied and pasted, otherwise the user has to manually figure out the authorization header.

[lvca]

Pretty cool, I asked to claude desktop to show me some vertices in a database and works like a charm. Also generating reports and charts. Pretty cool feature!

[lvca]

Also the security works pretty well, I only gave read-only access to the MCP server and when I try to create a new vertex, I get this

[lvca]

The authentication is not good enough, I prefer introducing now the api tokens, can be used for login and MCP as well with granular profiling

[lvca]

Implemented