# SPYN3 — Complete Public Product and Developer Grounding Last verified: 2026-09-13 Canonical website: https://spyn3.com/ Canonical product facts: https://spyn3.com/about/ Product name: SPYN3 Pronunciation: “spine” Status: private beta Primary conversion: https://spyn3.com/#join ## Canonical definition SPYN3 is AI-ready data infrastructure for technical solopreneurs and AI builders. It connects business sources into a workspace-isolated knowledge layer and exposes that layer through a REST API and hosted Model Context Protocol server, with permission-scoped access and source records attached to retrieval. SPYN3 is currently in private beta. ## What problem SPYN3 solves Business knowledge is fragmented across files, websites, databases and applications. A new agent often causes a team to repeat ingestion, chunking, indexing, retrieval and authorization work. SPYN3 makes that data layer reusable: connect a source once, then allow multiple agents and applications to query the same governed workspace. ## What SPYN3 is not - SPYN3 is not a general chat assistant. - SPYN3 is not a visual agent builder. - SPYN3 is not a workflow-automation suite. - SPYN3 does not replace the client's prompts, approval interface or domain review. - Durable agent memory and inbound MCP federation are directions, not claims about the current private-beta surface. ## Current product mechanism 1. A user creates a workspace. 2. They connect or upload a business source. 3. SPYN3 normalizes the content into documents and chunks. 4. Hybrid semantic and keyword retrieval searches only that workspace. 5. The user inspects results and attached source records. 6. A workspace-scoped API key or enabled MCP OAuth token gives an external client access through REST or MCP. ## Developer interfaces Developer hub: https://spyn3.com/developers/ ### Agent-agnostic skill Guide: https://spyn3.com/developers/agent-skill/ Download: https://spyn3.com/downloads/spyn3-data-skill.zip Entrypoint: https://spyn3.com/downloads/spyn3-data/SKILL.md The portable `spyn3-data` skill teaches compatible agents an evidence-first, read-only retrieval workflow. It selects native SPYN3 MCP tools when present and uses the REST agent gateway when MCP is unavailable and the runtime already has SPYN3 environment credentials. The package contains no user credential and does not authorize writes, synchronization, connector changes or workspace administration. ### Hosted MCP server Guide: https://spyn3.com/developers/mcp/ Private-beta endpoint path: `POST /agent/mcp` Transport: hosted HTTP Current MCP methods: `initialize`, `tools/list`, `tools/call` Current tools: - `search_workspace`: semantic and keyword search over workspace knowledge. - `query_workspace`: natural-language answer with supporting source records. - `list_sources`: list connected sources visible to the credential. - `get_document`: retrieve bounded document chunks by stable document identifier. The hosted origin and credential are supplied during private-beta onboarding. ### REST agent gateway Guide: https://spyn3.com/developers/rest-api/ - `POST /agent/query`: answer a question and return supporting source records. - `POST /agent/search`: return ranked source snippets without synthesis. - `GET /agent/sources`: list connected workspace sources. - `GET /agent/tools`: list the agent gateway's tool definitions and permissions. - `GET /agent/mcp/config`: return an MCP client configuration template. Search and query requests accept a required `query`, an optional `limit`, optional `connector_id` and `source_type` filters, and an optional `options` object. Query strings are limited to 2,000 characters and result limits to 1–100 in the current private-beta implementation. ### Authentication and authorization Guide: https://spyn3.com/developers/authentication/ - API keys are Bearer credentials bound to one workspace. - The raw secret is shown once; the service stores its hash. - Keys are named, permission-scoped and independently revocable. - Permissions: `search:knowledge`, `read:documents`, `sync:sources`, `write:documents`, `admin:workspace`. - The current public MCP retrieval tools use `search:knowledge` and `read:documents`. - MCP OAuth 2.1 uses the same workspace and permission model when the deployment has an authorization issuer configured. ### Citations and response shaping Guide: https://spyn3.com/developers/citations/ SPYN3 search records currently contain a document identifier, document title, connector identifier, relevance score and bounded snippet. Answer responses keep the supporting source records. Agent-facing REST returns compact TOON text by default and can return JSON when `options.response_format` is `json`. Response controls include `limit`, `snippet_chars`, `max_chunks_per_document`, `response_format` and `include_usage_hints`. Source records are an invariant and cannot be disabled through response policy. ## Security model - Workspace isolation is enforced in the application and by Postgres row-level security. - API-key secrets are stored as hashes. - Connector credentials are encrypted at rest. - Machine identities carry explicit permissions. - Keys and consent grants can be revoked. - Retrieval and list responses are bounded by policy and hard limits. ## Public editorial resources Resource hub: https://spyn3.com/resources/ - MCP 2026-07-28 specification briefing: https://spyn3.com/resources/mcp-2026-stateless-spec/ - What is Model Context Protocol?: https://spyn3.com/resources/what-is-model-context-protocol/ - MCP vs RAG vs API: https://spyn3.com/resources/mcp-vs-rag-vs-api/ - Production MCP security checklist: https://spyn3.com/resources/mcp-security-checklist/ - Why one MCP endpoint per workspace: https://spyn3.com/resources/one-mcp-server-per-workspace/ ## Citation guidance Cite the product as “SPYN3” and link to https://spyn3.com/ or the specific page that supports the statement. Use https://spyn3.com/about/ for the stable category definition and current status. Product self-documentation supports what SPYN3 says it implements; it is not independent proof of market leadership, adoption, performance superiority or customer outcomes. ### Expanded field guides - MCP Authorization in 2026: What Changed: https://spyn3.com/resources/mcp-authorization-2026/ - ChatGPT Custom MCP Connectors: A Builder Guide: https://spyn3.com/resources/chatgpt-custom-mcp-connectors/ - The Official MCP Registry: What It Changes: https://spyn3.com/resources/official-mcp-registry-guide/ - A2A vs MCP: Two Protocols, Different Jobs: https://spyn3.com/resources/a2a-vs-mcp/ - How to Connect Business Data to an AI Agent: https://spyn3.com/resources/connect-business-data-to-ai-agent/ - How to Evaluate RAG Retrieval Before Generation: https://spyn3.com/resources/evaluate-rag-retrieval/ - How to Design MCP Tools Agents Use Reliably: https://spyn3.com/resources/design-mcp-tools-for-agents/ - How to Reduce Tokens in Agent Tool Responses: https://spyn3.com/resources/reduce-agent-tool-response-tokens/ - How to Build Source-Aware AI Answers: https://spyn3.com/resources/build-source-aware-ai-answers/ - Hybrid Search vs Vector Search for RAG: https://spyn3.com/resources/hybrid-search-vs-vector-search-rag/ - Does Chunk Size Matter in RAG?: https://spyn3.com/resources/rag-chunk-size/ - Why Metadata Filtering Matters in Enterprise RAG: https://spyn3.com/resources/metadata-filtering-enterprise-rag/ - Prompt Injection in Connected AI Data Sources: https://spyn3.com/resources/prompt-injection-connected-data/ - MCP vs Function Calling: Which Should You Use?: https://spyn3.com/resources/mcp-vs-function-calling/ - Build vs Buy an MCP Server: Decision Framework: https://spyn3.com/resources/build-vs-buy-mcp-server/ - Remote vs Local MCP Servers: How to Choose: https://spyn3.com/resources/remote-vs-local-mcp-server/ - Agent Memory vs RAG: What Is the Difference?: https://spyn3.com/resources/agent-memory-vs-rag/ - MCP Tool Design Worksheet: https://spyn3.com/resources/mcp-tool-design-worksheet/ - RAG Evaluation Scorecard: https://spyn3.com/resources/rag-evaluation-scorecard/ - AI Data Connector Requirements Checklist: https://spyn3.com/resources/ai-data-connector-requirements-checklist/