
600+ connectors and 30k+ tools for AI agents, zero auth code
AgentKit by Scalekit is a connection and auth layer for AI agents: it exposes 600+ connectors and 30,000+ pre-built tools over MCP, handles per-user OAuth and credential injection at runtime, and lets you curate tool sets per agent via Virtual MCP servers. Your agent code calls the tools; Scalekit handles auth, retries, rate limits, and audit logging.
Pre-built connectors with ready-to-call tools schema-optimised for LLMs — not thin wrappers over REST endpoints. Auth, retries, and rate limits are handled on every call.
Tokens live in an isolated vault (AES-256 at rest, isolated per tenant) and are injected at runtime, so your agent calls the tool without ever touching the token or exposing credentials in code, logs, or LLM context.
Compose custom tool bundles from any connector, scoped to the role and the user. A smaller tool surface in context can cut token overhead by up to 80% and sharpen agent reasoning.
Define a schema and point it at any endpoint on an existing connector, or add internal APIs and custom MCP servers supporting OAuth 2.0, API key, RSA, and Basic auth — they then behave like built-in connectors.
Every tool call is tracked with success rate, error rate, and call volume per connector, with classification of connector vs API errors, structured event payloads, and log streaming to Datadog, Splunk, or any SIEM.
One config generates a per-user authenticated MCP URL that works with Claude, Cursor, Windsurf, and any MCP client, with framework integrations for Anthropic, OpenAI, Langchain, Google ADK, Mastra, and Vercel AI.
Cloud SaaS with US or EU data residency, private-cloud deployment inside your VPC, or fully self-hosted including air-gapped environments with BYO KMS and key policy.
Pick from the catalog of 600+ connectors, define a custom tool on an existing connector, or bring your own API/MCP server with supported auth schemes (OAuth 2.0, API key, RSA, Basic).
Each end user's credentials are stored in an isolated vault and injected at call time, so every tool call runs as the user the agent is acting for — no shared accounts or credential bleed.
Point any MCP client (Claude, Cursor, Windsurf, etc.) or agent framework (Langchain, Vercel AI, Google ADK, Mastra) at a generated per-user authenticated MCP URL, optionally curated into a Virtual MCP server with only the tools that task needs.
Monitor per-connector health, classify errors, and stream structured tool-call events to Datadog, Splunk, or any SIEM to see exactly what your agent ran and when.
Three plans: Free at $0/month with 10,000 tool calls and 10,000 trigger events per month, unlimited connected accounts, and the full connector catalog; Growth at $99/month with 100,000 tool calls and 50,000 trigger events included (plus overage of $0.50/1K calls and $3/1K events), custom connectors, branding, and EU data residency as a $99/mo add-on; Enterprise with custom volume, SSO/SCIM, BYOK,
$0/month
The complete platform, free: 10,000 tool calls/mo, 10,000 trigger events/mo, unlimited connected accounts, 600+ connectors and 30,000+ tools, bring your own credentials, community + email support.
$99/month
Everything in Free plus 100,000 tool calls/mo (+$0.50 per 1K overage), 50,000 trigger events/mo (+$3 per 1K overage), custom connectors & API proxy, customization & branding, basic dashboard roles, private Slack support, and EU data residen
Custom
Everything in Growth plus custom tool call & trigger volume, SSO/SCIM & custom roles, BYOK & IP allowlisting, SIEM log streaming & dashboard audit trail, HIPAA BAA & custom DPA, 99.99% SLA, VPC or on-prem deployment, forward-deployed engine
AgentKit is Scalekit's connection and auth layer for AI agents. It provides 600+ connectors and 30,000+ pre-built tools over MCP, handles per-user authentication and runtime credential injection, and adds observability so you write agent logic rather than auth and connectivity plumbing.
Every token lives in an isolated vault with AES-256 encryption at rest, isolated per tenant. Scalekit injects credentials at call time, so tokens never appear in your code, logs, or LLM context. Enterprise plans can bring their own GCP or AWS KMS for key rotation, revocation, and audit control.
Yes. The Free plan costs $0/month and includes the full connector catalog, unlimited connected accounts, and 10,000 tool calls plus 10,000 trigger events per month, with community and email support.
A Virtual MCP server is a curated bundle of tools pulled from any connector, scoped to the role and the user. Each agent gets only the tools it needs, which reduces what the model must reason over and can cut token overhead by up to 80%.
It works out of the box with Anthropic, OpenAI, Langchain, Google ADK, Mastra, and Vercel AI, and supports MCP clients including Claude, Cursor, and Windsurf — or any MCP client, via a per-user authenticated MCP URL generated from a single config.
Yes. You can define a custom tool schema pointed at any endpoint on an existing connector, or bring your own connector — an internal API or custom MCP server — with OAuth 2.0, API key, RSA, or Basic auth. Custom connectors and the API proxy are available on the Growth plan and above.
AgentKit tracks every tool call with per-connector health (success rate, error rate, call volume), classifies errors as connector vs API errors, captures structured tool-call event payloads, and streams logs to Datadog, Splunk, or any SIEM. SIEM log streaming and dashboard audit trails are Enterprise features.
Yes. Beyond the managed cloud (US or EU data residency), Enterprise offers deployment inside your VPC or fully self-hosted, including air-gapped environments, with BYO KMS, IP allowlisting, HIPAA BAA, and a 99.99% SLA.