Feature · BYOK + MCP + A2A
Open infrastructure. No vendor lock-in.
Bring your client's API keys. Connect 500+ tools via MCP. Speak A2A to other agent platforms. Three open standards, one platform.
What it does
BYOK: every sub-org can ship its own Anthropic / OpenAI / Gemini / Groq keys. Your client pays the model provider directly. KILN never marks up tokens — you keep 100% of the platform fee, and your client keeps full control of their AI spend.
MCP (Model Context Protocol): the open standard for tool calls. Connect any MCP server — there are 500+ in the community already covering Notion, Slack, Linear, Airtable, internal databases, custom HTTP services. KILN agents discover tools at runtime; no per-tool integration work.
A2A (Agent-to-Agent): Google's open standard for inter-agent communication. KILN agents can call and be called by any A2A-speaking agent — your sub-org's agent can hand off to a partner agency's specialized agent without leaving the protocol.
How It Works
Three moving parts
Connect provider keys (BYOK)
Per sub-org or per agent. Anthropic, OpenAI, Gemini, Groq, OpenRouter all supported. Override at the workflow node level for cost/performance routing.
Plug in MCP servers
Hosted MCP servers (Notion, Slack, GitHub, Linear) are one click. Custom MCP servers can be added by URL + bearer token. Tools surface in the workflow editor immediately.
Connect A2A agents
Register external A2A agents from your KILN dashboard. Workflows can call them with the same node interface as internal agents. Discovery via the public A2A directory.
Use Cases
What agencies actually build with this
Cost-controlled deployments
Use a cheap model (Haiku) for triage, route to Sonnet for reasoning, fall back to Gemini Flash for high-throughput batch. BYOK keeps the bill on the client.
Internal-tool access
Build an MCP server that exposes your client's CRM. Their agent now reads/writes CRM records natively, no glue code per client.
Cross-agency handoff
Your travel-booking agent calls a partner agency's flight-pricing agent via A2A. Both keep their own infrastructure; you split the revenue.
Technical details
- Per-org and per-agent API key management with RBAC enforcement
- Native MCP client — discovers tools, validates schemas, manages auth tokens
- MCP server registry with one-click installs from the marketplace
- A2A protocol implementation matching Google's spec (call, observe, stream)
- Per-call cost attribution: which provider, which key, which token cost
- Rate-limit + budget-cap enforcement before any external call fires
Ready to build with this?
Free forever for testing. Start charging your first client by next week.