Software bug in the database connection system

[KateBoher]

OS Linux Debian Bookworm
Cursor version 0.47.9
mcp.json:

{
  "mcpServers": {
    "pgLocal_postgres": {
      "command": "npx",
      "args": ["-y", "@smithery/cli@latest", "run", "@smithery-ai/postgres", "--config", "{\"databaseUrl\":\"postgresql://postgres:postgres@localhost:5432/postgres\"}"]
    },
    "pg79_sp_admin": {
      "command": "npx",
      "args": ["-y", "@smithery/cli@latest", "run", "@smithery-ai/postgres", "--config", "{\"databaseUrl\":\"postgresql://developer:[email protected]:5432/sp_admin\"}"]
    },
    "pg79_spo_online": {
      "command": "npx",
      "args": ["-y", "@smithery/cli@latest", "run", "@smithery-ai/postgres", "--config", "{\"databaseUrl\":\"postgresql://developer:[email protected]:5432/spo_online\"}"]
    },
    "pg79_vo_online": {
      "command": "npx",
      "args": ["-y", "@smithery/cli@latest", "run", "@smithery-ai/postgres", "--config", "{\"databaseUrl\":\"postgresql://developer:[email protected]:5432/vo_online\"}"]
    }
  }
}

All db connections are green.
Together with AI we came to the conclusion that this is a software error. This is what he wrote:

I discovered a software bug in the database connection system:

Bug Description:

1. Issue: The system ignores the requested connector name and always uses the configuration of the first connection from the list of available connectors

2. Expected behavior: The system should use the configuration of the connector that is explicitly requested by name

3. Actual behavior: The system only uses the configuration of the first connector in the list

4. Steps to reproduce:

• When calling mcp_pg79_spo_online_query, the system connects to the database whose configuration is first in the list

• When moving the desired connector to the beginning of the list, the system starts using its configuration

• When moving the connector to another position, its configuration is ignored, system uses the first connection in the list any way

This is a serious issue that can lead to:

• Connecting to the wrong database

• Using incorrect data schemas

• Potential data handling errors

I recommend registering this bug in the issue tracking system for further fixing.

---

We check the connection:
SELECT current_database();

[KateBoher]

The agent’s capabilities differ from the documentation. It cannot select a specific mcp tool. If I specify the required tool by name (pg79_spo_online), the agent selects mcp_pg79_spo_online_query (technical name?), but the chat simply displays just “query” and sql query is executed with the first tool in the list, not with the one I selected.

This is a bug description from Agent:

Bug Description: MCP Tool Name Inconsistency

1. Issue: There is a discrepancy between MCP server names in configuration and tool names provided to the agent

• In mcp.json, MCP servers are configured with clean names (e.g. pg79_spo_online)

• But the agent receives tools with technical names (e.g. mcp_pg79_spo_online_query)

1. Expected behavior:

• The agent should be able to use MCP server names as defined in configuration

• OR documentation should clearly state that tool names differ from MCP server configuration names

• The naming convention should be consistent across configuration and tool usage

1. Actual behavior:

• Agent sees tools with added prefixes (mcp_) and suffixes (_query)

• Cannot directly use MCP server names from configuration

• No clear mapping between configuration names and tool names

1. Impact:

• Confusion in tool selection and usage

• Potential errors when trying to use configuration names directly

• Inconsistent developer experience

1. Steps to reproduce:

• Define MCP server in config: pg79_spo_online

• Try to use this name directly: Error calling tool: Unknown tool: pg79_spo_online

• System only recognizes: mcp_pg79_spo_online_query

Reference: Cursor MCP Documentation

This appears to be a disconnect between the MCP configuration layer and the tool abstraction layer provided to the agent.