SPYN3 developer platform

One endpoint.Every system.

Connect your MCP servers, databases and APIs once. Then point Claude, Cursor or any MCP client at a single endpoint. Tools are foldered by source, namespaced, and scoped per key.

Connect an agent How federation works

mcp session

→ list_toolsets()
  github · 41 tools
  linear · 23 tools
  postgres · 3 tools

→ enable_toolset("linear")
  Enabled linear: 23 tool(s)

→ linear/list_issues(assignee: "me")
  …Linear's answer…

1 endpointEvery connected system behind one MCP URL per workspace.

44 presetsWell-known remote MCP servers, most signing in with OAuth.

Read / writeTwo permissions; a read-only key never sees a write tool.

Honest errorsA failing upstream is named, never hidden behind an empty result.

Choose your path / 01

Bring the client. Keep the connections.

Each guide answers one implementation job. The workspace, its connections and its permissions stay the same whichever client you use.

MCP quickstartPOST /agent/mcp

Connect an agent

Claude Code, Cursor or any remote-capable MCP client: one URL plus an API key, or OAuth.

Open MCP guide → Architecture<source>/<tool>

How federation works

Folders, namespaces, the tool catalog, read-only generated tools and honest errors.

Read the design → Contextspyn3://me/memories

Memory and skills

Shared memories, skills and notes for every agent, with Claude Code and Codex hooks.

Read the guide → Security modelBearer / OAuth

Scope every agent

API keys and MCP OAuth, read vs write permissions, per-source scope, encrypted upstream credentials.

Review authentication → HTTPJSON-RPC

Use it from a script

Discovery endpoints, and calling any connected tool with nothing but curl.

Open HTTP guide → Portable agent skillZIP / Markdown

Teach the agent the method

Open only what the task needs, read before acting, report failures with the source named.

Download the skill →

System shape / 02

Connect once. Change clients freely.

The agent is the replaceable part. Connections, credentials and permissions live in the workspace, so moving to a new client means moving one URL.

01 / SourcesMCP, Postgres, REST, Drive

Connected once per workspace.

02 / CatalogStored tool lists

A dead upstream keeps its tools listed.

03 / FoldFolders

Hidden until the agent opens one.

04 / BoundaryScope and log

Read/write, per-source, audited.

05 / ClientsAny MCP client

API key or OAuth.

Interface status

Generally available, with a free plan. The dashboard shows your endpoint, a one-time API key secret and copy-ready client configuration. Examples on these pages use $SPYN3_API_KEY, because a real key never belongs in a config file you commit.

Make the next agent a one-line change.

Connect the systems you already use once. Free covers 500 tool calls a month, no card.

Start free