Chroma MCP Server
Chroma MCP servers enable AI models to interact with the Chroma embedding database for vector search, collection management, and document operations.
Overview
The Chroma MCP Server connects AI models to Chroma, the open-source embedding database for building Python or JavaScript LLM apps with memory. It provides standardized tools to create and manage collections, add/query documents via vector search and full-text search, and filter by metadata. Embedding throughput is often the real bottleneck at scale — Manticore Search's ONNX work shows how much headroom there is.
Created by:
Developed by Chroma
Key Features
Flexible Client Types
Ephemeral, persistent (file-based), HTTP for self-hosted, and Cloud client
Collection Management
Create, modify, delete; list with pagination; HNSW configuration
Document Operations
Add, query, get, update, delete; metadata filtering and full text search
Embedding Functions
Supports default, Cohere, OpenAI, Jina, VoyageAI, Roboflow; persists per-collection
Available Tools
Quick Reference
| Tool | Purpose | Category |
|---|---|---|
chroma_list_collections | List collections with pagination | Discovery |
chroma_create_collection | Create a new collection | Schema |
chroma_modify_collection | Update name or metadata | Schema |
chroma_delete_collection | Delete a collection | Schema |
chroma_add_documents | Insert documents with metadata/IDs | Write |
chroma_query_documents | Semantic query with filters | Read |
chroma_get_documents | Retrieve by IDs/filters | Read |
chroma_update_documents | Update content/metadata/embeddings | Write |
chroma_delete_documents | Remove documents | Write |
Detailed Usage
chroma_list_collections▶
List all collections with optional pagination parameters.
use_mcp_tool({
server_name: "chroma",
tool_name: "chroma_list_collections",
arguments: {
offset: 0,
limit: 20
}
});
chroma_create_collection▶
Create a collection with optional metadata and HNSW configuration.
use_mcp_tool({
server_name: "chroma",
tool_name: "chroma_create_collection",
arguments: {
name: "my_collection",
metadata: { project: "docs" },
hnsw: { m: 16, efConstruction: 200 }
}
});
chroma_add_documents▶
Add documents with optional metadata and custom IDs.
use_mcp_tool({
server_name: "chroma",
tool_name: "chroma_add_documents",
arguments: {
collection: "my_collection",
ids: ["doc-1", "doc-2"],
documents: [
"First document text",
"Second document text"
],
metadatas: [
{ source: "notes" },
{ source: "manual" }
]
}
});
chroma_query_documents▶
Query documents using semantic search with advanced metadata filtering.
use_mcp_tool({
server_name: "chroma",
tool_name: "chroma_query_documents",
arguments: {
collection: "my_collection",
query_texts: ["pagination implementation"],
n_results: 5,
where: { source: "manual" }
}
});
chroma_get_documents▶
Retrieve documents by IDs or metadata filters with pagination.
use_mcp_tool({
server_name: "chroma",
tool_name: "chroma_get_documents",
arguments: {
collection: "my_collection",
ids: ["doc-1"],
offset: 0,
limit: 10
}
});
Installation
{
"mcpServers": {
"chroma": {
"command": "uvx",
"args": [
"chroma-mcp",
"--client-type",
"ephemeral"
]
}
}
}
Persistent / Cloud / HTTP:
Configure clients via arguments:
Persistent: "args": ["chroma-mcp", "--client-type", "persistent", "--data-dir", "/path/to/data"]
Cloud: "args": ["chroma-mcp", "--client-type", "cloud", "--tenant", "your-tenant", "--database", "your-db", "--api-key", "your-api-key"]
HTTP: "args": ["chroma-mcp", "--client-type", "http", "--host", "your-host", "--port", "your-port", "--custom-auth-credentials", "your-credentials", "--ssl", "true"]
Embedding Function API Keys:
When using external embedding providers (Cohere, OpenAI, Jina, VoyageAI, Roboflow), set the corresponding API key environment variables per provider. Embedding function persistence requires Chroma v1.0.0+.
For secure local setup, pass --dotenv-path to load keys from a custom .env file.
Example: "args": ["chroma-mcp", "--dotenv-path", "/custom/path/.env"]
Sources
Related Articles & Guides
Milvus MCP Server
Connect AI assistants to Milvus, the open-source vector database, for semantic search, vector storage, and hybrid query operations through the Model Context Protocol.
WhoDB MCP Server
Connect AI assistants to WhoDB, the all-in-one database management tool, for querying and exploring PostgreSQL, MySQL, SQLite, MongoDB, Redis, and more through MCP.
Vectara MCP Server
Vectara MCP servers enable AI models to interact with Vectara's RAG platform, providing capabilities for fast and reliable Retrieval-Augmented Generation, semantic search, and hallucination correction.