toolbelt.ai

What It Is

Toolbelt is an “agentic data layer” built by Kinetica, positioned as a collaborative data foundation for AI agents. It does not directly claim to provide its own proprietary large language model. Instead, through a single HTTP MCP endpoint, it exposes capabilities such as SQL queries, vector search, knowledge graph traversal, schema introspection, and ingestion-job inspection as toolbelt_* tools for MCP-compatible clients including Claude, Cursor, Windsurf, ChatGPT, Gemini, Codex CLI, and OpenClaw.

Core Capabilities and Integrations

Its biggest value lies in its unified access layer: developers can install it with a single npx @toolbeltai/cli command, after which the CLI automatically detects AI clients, writes MCP configuration, and generates credentials. Claude Code, Cursor, Windsurf, and similar tools can use native HTTP MCP; Claude Desktop, which only natively supports stdio servers, requires the mcp-remote shim. For team use cases, Toolbelt supports namespace isolation by user or team, with assets, vectors, and graphs separated by namespace UUIDs, making it easier to manage multiple projects and teams.

Pricing and Value

The free tier is fairly clear: anonymous users get 1,000 calls per month and 1 namespace; after email verification, the free plan includes 2,000 calls per month, 1GB of storage, and up to 10 namespaces. Pro costs USD 29/month and includes 150,000 calls, 50GB, and 50 namespaces; Team costs USD 89/month and includes 500,000 calls, 100GB, and unlimited namespaces. The Enterprise plan offers Self-Hosted deployment, Helm chart, and commercial support, but requires contacting sales, and the Helm chart is marked as private beta. For MCP developers, the entry cost is low, while the Pro and Team call quotas are also fairly attractive.

Pros, Cons, and Limitations

The strengths are its short installation path, coverage of mainstream AI coding and agent clients, tool types that closely match the needs of data-centric agents, and its Apache 2.0 open-core approach. The limitations are also clear: the main materials do not disclose the underlying model, output accuracy, data encryption, audit capabilities, compliance certifications, or data retention policy; Chinese UI, Chinese documentation, and payment methods are also not specified. Since it is primarily a tool layer, the final answer quality still depends on the connected model and the quality of the data.

Who It Is For and Access from China

Toolbelt is suitable for developers and data teams already using tools such as Claude Code, Cursor, and Windsurf, and who want agents to access databases, vector stores, or knowledge graphs safely. The main materials do not provide details on access from mainland China, and payment methods are unknown. If network access or payment is limited, alternatives include self-hosting an MCP Server, using domestic vector database/knowledge base solutions, or choosing an agent data-layer alternative that supports local deployment.