Coredence

What Coredence does

The Coredence MCP server lets an AI assistant query an identity that the Coredence desktop app has already built from your written work. That identity describes patterns in how you work, including rules, the evidence supporting those rules, the subjects they concern, and how you make choices when no explicit guidance exists.

The server is intended for assistants that need to ask questions about this identity rather than receive the underlying database file. It does not create or revise the identity through MCP. Its interface is read-only, so an assistant can retrieve and quote the stored material but cannot decide whether the identity is correct or change it.

Coredence belongs in the Knowledge & Memory category because it gives an assistant access to persistent, evidence-linked information about a person’s working patterns. The identity remains in a single local file created by the desktop application.

How it works

The Coredence MCP server runs on the same machine as the application and communicates with the MCP client through a local pipe. The client starts the coredence command, which opens the identity database in read-only mode. There is no account, API key, or listening network port involved in this connection.

The database is opened in a way that prevents writes at the SQLite layer. This makes the read-only behavior an operational property of the connection, not merely a convention for clients to follow. The assistant can inspect the identity and challenge it, while changes remain under the user’s control in the Coredence application.

The engine that initially derives the identity is separate from this repository. During that distillation process, it can call a selected Claude Code setup, a user-provided API key, or an OpenAI-compatible server, including one running locally. Those model calls are part of building the identity, not of the MCP server’s read-only query path.

Setup and configuration

Download and install Coredence first, then launch it once so the coredence command is available. The application is distributed from the project’s releases page and supports macOS 14 or later on Apple silicon; the README also states that Windows is supported. The identity is stored as one file and can move between machines with you.

The project provides a Claude Desktop desktop extension in .mcpb format. It also supplies plugin and extension configuration for Claude Code, Codex, Gemini CLI, VS Code, and GitHub Copilot. For other supported clients, the command coredence connect <client> can write configuration for Claude Code, Cursor, VS Code, Windsurf, JetBrains, Codex, or Gemini CLI.

Tools and capabilities

The Coredence MCP server exposes five read-only tools:

• identity returns the complete identity by layer.
• rules finds what the person does or values about a subject.
• evidence retrieves the concrete moments supporting a rule.
• axes describes how the person chooses when no explicit statement exists.
• topics lists the subjects associated with the identity.

Together, these tools let an assistant move from a broad identity query to a specific rule and then to the evidence behind it. They are useful when responses should be informed by observed work rather than by an unsupported profile or a copy of the database.

Limitations and notes

The Coredence MCP server only answers questions about an identity that the desktop app has already generated. It does not include the private engine used to read source material and distil rules, and the MCP interface does not provide mutation tools.

The server is local and read-only, but the earlier distillation stage may contact the model selected during setup. Choosing a server on the user’s own machine can avoid sending that material elsewhere. The repository is a distribution, documentation, and issue-tracking project rather than the source for the identity engine. The README identifies the MCP Registry entry as io.github.coredence/coredence.