MCP overview

Use Satellyte through an agent-facing MCP layer aligned with the Public API.

The SatellyteAI MCP Server lets MCP-compatible clients and AI agents use Satellyte capabilities through intent-shaped tools, reusable resources, and guided prompts.

It is aligned with the Satellyte Public API, but it is not a one-to-one mirror of every REST endpoint. The MCP surface is intentionally capability-based so agents can use fewer, clearer tools for full GTM workflows.

What MCP adds

With the MCP server, an agent can:

Discover target companies and people.
Resolve company identity between domains and LinkedIn URLs.
Pull company signals such as hiring, funding, tech stack, financing events, and news.
Find decision makers, buying committees, employees, or waterfall ICP contacts.
Enrich contact records with email and/or phone data.
Work with radars, leads, exports, and outreach feedback.
Use guided workflow prompts for signal-based prospecting and account research.

How it relates to the Public API

Layer	Purpose
Public API	The HTTPS JSON API under `/api/v1`. It remains the source of truth for endpoint behaviour, workspace scoping, authentication, and credit charging.
MCP server	The agent-facing layer. It groups Public API capabilities into tools, resources, and prompts that are easier for AI clients to use safely.
MCP client	The tool host used by the user or agent, such as Claude Code or another MCP-compatible client.

Current status

The current MCP implementation includes:

STDIO transport for local Claude Code development.
HTTP transport for local MCP server testing.
JWT auth mode for local HTTP testing.
/health and /mcp endpoints in HTTP mode.
Satellyte app connection flow for getting the MCP Server URL and session token.
Agent-facing MCP tools, resources, and prompts.
Credit metadata and live BE-derived credit state where available.
dry_run support for charged or credit-sensitive flows where implemented.

Production-grade remote/public MCP deployment hardening is still a planned next-phase item. Do not treat local development defaults as a production security model.

Key design rule

MCP coverage is capability-based. A single MCP tool can cover multiple Public API routes when that gives agents a safer and clearer interface.

For example, get_company_signals consolidates multiple signal routes, and find_contacts consolidates multiple contact discovery strategies.

Where to go next

Setup and local validation: configure and smoke test the MCP server.
Authentication: connect a Satellyte API key safely.
Tools, resources, and prompts: inspect the current MCP surface.