Context Portal (ConPort) is a memory bank server designed to build project-specific knowledge graphs for AI assistants, enabling context-aware development within IDEs. It targets developers and AI agents seeking to enhance RAG capabilities with structured project data, offering a more efficient and queryable alternative to file-based context management.

How It Works

ConPort acts as a Model Context Protocol (MCP) server, utilizing a SQLite database per workspace to store structured project information. It builds a knowledge graph by capturing entities like decisions, progress, and architecture, along with their relationships. Vector embeddings are integrated for semantic search, powering advanced RAG. The server is deployed via STDIO for tight IDE integration, with explicit workspace_id management for multi-workspace support.

Quick Start & Requirements

• Install/Run: Recommended: uvx --from context-portal-mcp conport-mcp --mode stdio --workspace_id ${workspaceFolder}. Developer install: git clone and uv pip install -r requirements.txt.
• Prerequisites: Python 3.8+, uv (recommended).
• Setup: Minimal setup time, primarily configuration within IDE MCP settings.
• Docs: conport_mcp_deep_dive.md

Highlighted Details

• Enables building dynamic project knowledge graphs with explicit item relationships.
• Integrates vector data storage and semantic search for advanced RAG.
• Provides structured context for AI assistants, supporting prompt caching.
• Offers a comprehensive suite of MCP tools for managing product context, decisions, progress, system patterns, and custom data.

Maintenance & Community

• Active development with clear contribution guidelines.
• Links to community channels or developer resources are not explicitly provided in the README.

Licensing & Compatibility

• Licensed under Apache-2.0.
• Compatible with commercial use and closed-source linking.

Limitations & Caveats

The README does not explicitly detail community support channels (e.g., Discord, Slack). While it mentions custom instruction files for specific IDEs, broader platform compatibility beyond MCP clients is not detailed.