generated: '2026-08-04'
method: searched
status: published
source: https://www.kinetica.com/agents
server:
  name: toolbelt
  product: Toolbelt
  vendor: Kinetica
  transport: http
  url: https://mcp.toolbelt.ai/mcp
  site: https://www.toolbelt.ai
  overview: https://www.kinetica.com/agents
  install: npx @toolbeltai/cli
  self_host: Helm chart (self-host the whole stack on your own cluster) or the hosted
    edition
authentication:
  type: oauth2
  gated: true
  observed: HTTP 401 with WWW-Authenticate:Bearer resource_metadata="https://mcp.toolbelt.ai/.well-known/oauth-protected-resource/mcp"
  protected_resource_metadata: https://mcp.toolbelt.ai/.well-known/oauth-protected-resource
  authorization_server: https://app.toolbelt.ai
  authorization_endpoint: https://app.toolbelt.ai/oauth/authorize
  token_endpoint: https://app.toolbelt.ai/oauth/token
  registration_endpoint: https://app.toolbelt.ai/oauth/register
  scopes:
  - mcp:tools
  - openid
  pkce: S256
  grant_types:
  - authorization_code
  - refresh_token
  bearer_methods:
  - header
tools_source: https://www.kinetica.com/llms.txt and https://www.kinetica.com/agents
  — names and capability descriptions only. The live tools/list is OAuth-gated, so
  per-tool inputSchema requires authenticated introspection and is NOT recorded here.
tools:
- name: toolbelt_sql
  description: Execute SQL against the connected Kinetica database.
  schema: requires authenticated introspection
- name: toolbelt_vector_search
  description: Vector similarity search over embeddings stored alongside structured
    and geospatial data.
  schema: requires authenticated introspection
- name: toolbelt_graph_traversal
  description: Traverse the knowledge graph / property graph in the connected database.
  schema: requires authenticated introspection
- name: toolbelt_schema_introspection
  description: Introspect schemas, tables and columns available to the agent.
  schema: requires authenticated introspection
- name: toolbelt_ingestion_job_inspection
  description: Inspect the state of streaming/batch ingest jobs.
  schema: requires authenticated introspection
features:
- namespace isolation between datasets exposed to an agent
- one endpoint for any MCP-speaking client (Claude, Cursor, Windsurf, ChatGPT, Gemini,
  Codex CLI)
related_servers:
- name: mcp-kinetica
  transport: stdio
  kind: open-source, self-run
  package: https://pypi.org/project/mcp-kinetica/
  repository: https://github.com/kineticadb/mcp-kinetica
  note: first-party Python MCP server for a Kinetica database, published by kineticadb
- name: Kinetica Docs MCP
  transport: http
  kind: documentation, anonymous
  manifest: https://docs.kinetica.com/.well-known/mcp.json
  file: kinetica-docs-mcp.yml
x-evidence:
  fetched: '2026-08-04'
  probes:
  - url: https://mcp.toolbelt.ai/mcp
    method: POST tools/list
    http_status: 401
    body: '{"error":"unauthorized","error_description":"Bearer token required"}'
  - url: https://mcp.toolbelt.ai/.well-known/oauth-protected-resource
    http_status: 200
    content_type: application/json
  - url: https://app.toolbelt.ai/.well-known/oauth-authorization-server
    http_status: 200
    content_type: application/json