Your agent connects to a Corebee org, ingests the customer docs, brands the widget, and runs support end to end through one MCP and REST API. No dashboard. No clicks.
# resolve the org this key is scoped to
GET /api/v1/me
200 OK
{
"organization": {
"id": "org_a1c8...",
"name": "Acme Inc"
}
}Machine-readable, above the fold
The same workspace, whether your agent speaks plain REST or the Model Context Protocol. Real endpoints, real tool names, no inferred call order.
# 1. Verify the key and resolve the org it is scoped tocurl https://corebee.ai/api/v1/me \ -H "Authorization: Bearer $COREBEE_API_KEY"# 200 -> { "organization": { "id": "...", "name": "Acme Inc" } }# 2. Create a contact for the customer's end-usercurl -X POST https://corebee.ai/api/v1/contacts \ -H "Authorization: Bearer $COREBEE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "email": "buyer@acme.com", "name": "Dana Buyer" }'# 3. Read the support conversations the AI is handling# (status is an integer code; see the OpenAPI spec for the enum)curl "https://corebee.ai/api/v1/conversations?limit=20" \ -H "Authorization: Bearer $COREBEE_API_KEY"// Same workspace, over the Model Context Protocol.// Connect: https://corebee.ai/api/v1/mcp/stream (OAuth 2.1, scoped per org)// Ingest the customer's docs into the knowledge baseawait mcp.callTool("add_knowledge_url", { url: "https://acme.com/docs",});// Brand the widget for the customerawait mcp.callTool("configure_widget_appearance", { widget_color: "#2E7D6B", position: "right",});// Read how the AI support is performingconst kpis = await mcp.callTool("get_metrics", {});Every artifact is static, discoverable, and points to the next. Discovery leads to the contract, the contract leads to the tools.
https://corebee.ai/llms.txt
Discovery breadcrumb. Brand facts plus pointers to the API base URL, OpenAPI spec, agent manifest, and MCP endpoint.
https://corebee.ai/.well-known/agents.json
Links plus the connect, ingest-KB, configure-widget, operate flow, so the agent does not have to infer the call order.
https://corebee.ai/.well-known/mcp/server-card.json
Points at the live Corebee MCP endpoint and its OAuth protected-resource metadata. Tools are typed and self-describing.
https://corebee.ai/.well-known/oauth-protected-resource
RFC 9728 discovery. Leads to the authorization server, Dynamic Client Registration, and per-org scoped tokens.
https://corebee.ai/api/v1/openapi.json
The v1 REST contract: typed request and response schemas for conversations, contacts, labels, and webhooks.
https://corebee.ai/api/v1/mcp/stream
The live Model Context Protocol server. Connect any MCP-capable agent and call 40 typed Corebee tools directly.
Five steps, no human in the loop. Discover, authenticate, connect, ingest, deploy.
The agent fetches llms.txt, the agent manifest, and the MCP server card to learn the endpoints, flows, and auth requirements. No human reads docs.
OAuth 2.1 authorization code with PKCE, self-registered through Dynamic Client Registration, or a scoped API key. The agent gets a token bound to a single org.
Verify the credential and resolve the workspace with GET /api/v1/me. Every call from here is scoped to that one customer org. Nothing leaks across tenants.
Point Corebee at the customer docs URL with add_knowledge_url. RAG indexing builds the answer base, which resolves about 86% of questions automatically.
Brand the widget with configure_widget_appearance, then read conversations and metrics and reply, all through the API. A live AI support widget is answering, no dashboard touched.
Live today
Everything above runs against a connected Corebee org with no UI: ingest the knowledge base, brand the widget, read conversations and metrics, and reply, all through the REST v1 API and the MCP server. The org is created once through signup or our partner flow.
Early access
Fully autonomous account provisioning, where your agent creates a brand-new customer org and subscribes it to the flat $99/mo plan without a human, is in early access. If you are building that, tell us about your use case and we will get you set up.
An API-first help desk built for the agent doing the work, not the human who never logs in.
A live Model Context Protocol server, discoverable through well-known metadata. Connect any MCP-capable agent and call typed tools directly.
The agent manifest bundles the connect, ingest, and deploy sequence with flows and links so your agent succeeds on the first try.
Calls hit the same live org your customer uses. Ingest docs, brand the widget, and reply for real, with a widget answering customers in minutes.
Flat $99/mo, unlimited resolutions, seats, and conversations. No per-seat math and no overage surprises for your agent to model.
RAG retrieval over the ingested docs resolves the bulk of questions with no tuning. Your agent ships support that actually works.
Every action is attributed. You can tell exactly what your agent did versus what a person did, with a full trail per org.
Letting an agent run support is only safe when the blast radius is bounded. Corebee scopes it on every axis.
A representative slice of the 40 typed tools. Each tool declares its access level so your agent can reason about scope before it calls.
add_knowledge_urlwriteCrawl and index a docs URL into the RAG knowledge base.search_knowledgereadQuery the indexed knowledge base and return matching chunks.configure_widget_appearancewriteSet widget color, logo, position, and welcome text.list_conversationsreadList support conversations and their AI resolution status.reply_to_conversationwriteSend a reply to a conversation, visible to the customer.get_metricsreadAuto-resolution rate, volume, response time, and deflection.create_contactwriteCreate or update an end-user contact record.delete_knowledge_sourcedestructiveRemove a source and its indexed content. Requires confirmation.End to end, for its own customer. No human in the loop.
FAQ
Read the docs, wire up the flow, and let your agent onboard customers end to end. We are happy to help you design the integration.