MCP

Saaya, in every model you use.

The Model Context Protocol endpoint lets Claude, ChatGPT, Cursor, and any MCP-compatible client query your Saaya workspace directly. No glue code, no shims.

What is MCP

A wire between your model and your platform.

MCP is an open protocol any AI client can speak. Saaya exposes one MCP endpoint at https://mcp.saaya.ai — your client authenticates, lists the tools you've granted it, and starts calling them mid-conversation.

The model is yours. The conversation is yours. Saaya just becomes one of the things it can reach for.

Endpoint

https://mcp.saaya.ai

Auth

Bearer token, same key as the REST API. Permissions inherit from the key's RBAC role.

Default scope

Read-only by default. Write tools can be enabled per-key from the dashboard.

Why use it

Your platform, inside your prompts.

Summarize last week's sessions

"Claude, pull the last 30 sessions and summarize the top three themes customers raised."

Draft prompt iterations from data

"ChatGPT, look at my top-performing inbound qualifier and write three variants tuned for healthcare."

Query KBs from inside Cursor

"Cursor, search the Saaya knowledge base for the latest pricing tier — I'm wiring it into my checkout."

Surface campaign signals in one prompt

"Surface the bottom-quartile campaigns by answer rate and tell me what's common across them."

Connect

Hook Saaya into your client.

Claude Desktop

Add the Saaya server to your Claude Desktop config. Restart Claude — the tools light up automatically.

claude_desktop_config.json
{
  "mcpServers": {
    "saaya": {
      "url": "https://mcp.saaya.ai",
      "headers": {
        "Authorization": "Bearer ${SAAYA_API_KEY}"
      }
    }
  }
}

Cursor

Drop the Saaya bridge into your project's .cursor/mcp.json. The bridge handles the HTTP stream so Cursor sees a local MCP process.

.cursor/mcp.json
// .cursor/mcp.json
{
  "mcpServers": {
    "saaya": {
      "command": "npx",
      "args": ["@saaya/mcp-bridge"],
      "env": { "SAAYA_API_KEY": "sk_live_..." }
    }
  }
}
Continue.dev, Zed, Cline, and any other MCP-compatible client follow the same pattern — point them at https://mcp.saaya.ai with a Bearer token. Saaya speaks the standard.
Toolset

The tools Saaya exposes.

ToolScopeInputOutput
  • list_agents

    Return all agents in the workspace, including drafts and published versions.

    read{ status?: "draft" | "published", limit?: number }{ agents: Agent[], next_cursor?: string }
  • get_agent

    Fetch a single agent by ID with full config and version history.

    read{ agent_id: string, include_history?: boolean }{ agent: Agent, history?: AgentVersion[] }
  • query_knowledge_base

    Run a semantic + keyword query against a knowledge base. Returns chunks with citations.

    read{ kb_id: string, query: string, top_k?: number }{ chunks: Chunk[], confidence: number }
  • list_sessions

    Return recent sessions, optionally filtered by agent, channel, or sentiment.

    read{ agent_id?: string, channel?: Channel, since?: string }{ sessions: SessionSummary[] }
  • get_session_transcript

    Fetch the full transcript for a session, including tool calls and timing.

    read{ session_id: string }{ transcript: Turn[], duration_ms: number }
  • get_session_summary

    AI-generated summary of a session: outcome, sentiment, and follow-ups.

    read{ session_id: string }{ summary: string, outcome: string, sentiment: number }
  • list_campaigns

    Return campaigns with high-level stats and current run state.

    read{ status?: CampaignStatus, limit?: number }{ campaigns: CampaignSummary[] }
  • get_campaign_status

    Live status for a single campaign — progress, success rate, fallback usage.

    read{ campaign_id: string }{ status: CampaignStatus, stats: CampaignStats }

Security

  • RBAC permissions are enforced per API key — every MCP call inherits the key's role.
  • Default scope is read-only. Write tools (create / publish / start) require explicit opt-in per key.
  • Every tool call lands in the audit log with the key ID, tool name, params, and trace ID.
  • HMAC-signed responses on streamable channels — clients verify before applying.
  • Saaya never forwards your data to a third-party model on its own.
MCP FAQ

Common questions

Plug it in

Bring Saaya into the model.

Generate a key, drop the URL into your client, and your sessions, KBs, and campaigns are one prompt away.

Get an API keyRead the API reference

No credit card · 15-min setup · Cancel anytime