chroma -custom

This MCP server is a Python-based implementation of the Model Context Protocol (MCP) tailored for the Chroma embedding database. It exposes a set of tools for managing Chroma collections, documents, and the enhanced sequential-thinking workspace, all via an MCP-compliant HTTP server (FastMCP) running on port 10550. You can list, create, modify, and delete collections; add and query documents with semantic search and advanced metadata filtering; and utilize the sequential thinking tools to model multi-step reasoning with revisions, branches, and session histories. The server leverages Chroma as its data backend and provides a rich API surface that mirrors the MCP tool namespace, including tools such as chroma_list_collections, chroma_create_collection, chroma_query_documents, chroma_add_documents, chroma_get_documents, and the sequential thinking family (chroma_sequential_thinking, chroma_get_similar_sessions, chroma_get_thought_history). To start, run the server via the uvx-based command, then issue MCP tool invocations to interact with Chroma-backed collections and documents. If you’re building an LLM workflow, you can orchestrate prompts, store generated data as documents, and retrieve results through semantic search and metadata filters.

Prerequisites

• Python 3.11+ (recommended) and a Python development environment
• Node.js is not required for this Python MCP server
• Internet access to install dependencies

Installation steps

1. Install the uv runtime for Python environments (or ensure your environment supports uv):

   pip install uv

2. Install the Chroma MCP server package (assuming the forked package is named chroma-mcp and available on PyPI or via your internal index):

   pip install chroma-mcp

   If you are using a local/alternate distribution, install according to your package manager (e.g., python -m pip install /path/to/chroma-mcp).

3. Run the MCP server using the uvx entry point:

   uvx chroma-mcp

   Optional environment variables you may want to set:

   • CHROMA_MCP_LOG_LEVEL: INFO|DEBUG to control logging verbosity
   • CHROMA_MCP_PORT: port number (default 10550 for FastMCP HTTP transport)

4. Verify the server is listening on port 10550 and accessible via MCP tool invocations. If you’re containerizing, you can adapt the uvx command inside your Dockerfile or docker-compose setup accordingly.

Notes and tips:

• Default transport port for the MCP HTTP server is 10550; ensure this port is open in your firewall or container networking.
• The server uses a Chroma-backed storage layer; you can create and manage collections, add documents, and perform semantic queries with metadata filters.
• For production deployments, consider using the Docker-first deployment pattern described in the fork’s docs, and enable security headers (e.g., Origin checks) to prevent unauthorized cross-origin access.
• If you upgrade MCP SDK versions, verify backward compatibility of existing tool definitions; this fork explicitly notes backward compatibility with optional new features when clients support them.
• Environment variables can be used to fine-tune logging, transport behavior, and port bindings; consult the project documentation for a full list of supported settings.