# One context layer. Every kind of agent.

Agents are becoming the way we work. Synorb gives them live, source-grounded context so they can reason with clarity, provenance, and follow-through.

## Shift from retrieving to receiving.

Enable your agent to receive information without burning run-time tokens on open web search.

### For Building **Coding**
Build with Cursor, Replit, Lovable, and more.

### For Tasking **Personal**
Task with Instinct, Muse, Town, and more.

### For Orchestrating **Horizontal**
Orchestrate with n8n, Flowise, Make.com, and more.

### For Working **Computer**
Work with CoWork, Operator, Computer and more.

### For Talking **Voice**
Talk with ElevenLabs, Vapi, Hume, and more.

## One Synorb Key

The same Synorb Key authenticates MCP and REST. Keep it only in the host or server secret store; it never belongs in browser code, a repository, logs, or a prompt.

```
Authorization: Bearer YOUR_SYNORB_KEY
```

```
{
  "mcpServers": {
    "synorb": {
      "url": "https://mcp.synorb.com/mcp",
      "headers": {"Authorization":"Bearer YOUR_SYNORB_KEY"}
    }
  }
}
```

## Agent operating sequence

**Resolve.** Use `GET /ontology/tags` before a tag filter. A guessed term can quietly return no matches.

**Scope.** Use catalog and details to choose explicit `stream_ids`. Omit it or pass `stream_ids: null` only for a deliberately broad scope; an explicit empty list returns `empty_stream_ids` without quota instead of becoming global.

**Retrieve.** Use MCP for an interactive loop. For REST, call `/manifests/plan`, then replay its request exactly to `/manifests/exec` with only `plan_token` added.

**Present.** Preserve returned source citations beside factual claims. A web product calls REST from its server-side route.

## Beacons and quotas

Use a Beacon for a recurring watch, direct retrieval for a one-off question. Convert a user intent into canonical `stream_ids` and `topic_filters`; intent text itself is descriptive, not executable. Poll with `GET /streams/{id}/manifests?beacon_id=...`.

Catalog, ontology, count/orient, planning, and Beacon CRUD use zero Manifest quota. Executed content bills only returned on-topic Manifests and does not double-bill duplicates in the dedup window. Fully unresolved exact tag names return `tag_names_unresolved` with `unresolved_tag_names`; mixed names use only the resolved scope for `tag_logic=or`, while `tag_logic=and` never drops an unknown required tag.

## Read the runtime sources

[**Agent guide** Five recipes, request/response pairs, billing, and failure contracts.](/content/agent-guide.md)  
[**llms.txt** Compact bootstrap for runtime loading.](/content/llms.txt)  
[**OpenAPI** Generated REST types and operations.](/content/openapi.json)

## For coding agents

Use MCP while the agent investigates coverage; use server-side REST in the product it ships. The build guides include the secure live-content pattern and citation requirement.

[**Lovable** Build cited user experiences with server-side REST.](/content/agents/coding-agents/lovable-mcp/index.html)  
[**Replit** Put the one key in a deployment secret and call from the backend.](/content/agents/coding-agents/replit-mcp/index.html)  
[**Cursor** Explore by MCP, ship by REST.](/content/agents/coding-agents/cursor-mcp/index.html)