Getting Started
TokST keeps durable knowledge available to authorized people and Agents. Start with Cloud MCP for memory work. Add the local Runtime only when this device will connect Agents, capture Sessions, or execute Tasks.
1. Connect Cloud MCP
Add the public endpoint to an MCP client that supports Streamable HTTP:
{
"mcpServers": {
"tokst": {
"type": "streamable-http",
"url": "https://api.tokst.com/mcp"
}
}
}
Complete the browser approval opened by your MCP client. For a static or headless client, create a dedicated key in Dashboard API Keys and store it in the client's protected credential settings:
{
"mcpServers": {
"tokst": {
"type": "streamable-http",
"url": "https://api.tokst.com/mcp",
"headers": {
"Authorization": "Bearer tk_live_your_api_key"
}
}
}
}
Use the MCP Server Reference for client-specific setup and authorization recovery.
2. Verify the Connection
Ask the connected Agent:
Search TokST for the current deployment policy. Return the matching memory title and ID.
The Agent uses tokst_search. A returned memory title and ID confirm that the
connection, authorization, Workspace, and Atlas route are ready.
3. Save Durable Knowledge
Ask the Agent to write confirmed information in Markdown:
Save this confirmed decision as a TokST decision memory in Markdown:
# Production release approval
- Production releases require approval from the Workspace owner.
- This policy applies to every production deployment.
The Agent uses tokst_remember. Open Dashboard -> Memories to review,
edit, archive, and trace the saved record.
Store facts, decisions, preferences, rules, and architecture that remain useful across future work. Keep passwords, API keys, private keys, raw logs, and temporary tool output outside regular memories. A Workspace Owner can enable Controlled original storage in Workspace settings for a deliberate, audited sensitive-memory write.
4. Give the Agent the Skill
Send this to an Agent that supports Skills or can read web pages:
Read https://tokst.com/skill.md. Connect to TokST Cloud MCP at https://api.tokst.com/mcp, complete authorization, then use MCP tools for memory search, context, and durable Markdown memory writes.
The Skill directs the Agent to search before repeating work, preserve the working language, and retain confirmed long-term knowledge without credentials or raw execution noise.
Advanced: Sessions, Memories, and Tasks
| Object | Use it for |
|---|---|
| Session | Capture normal Agent work, results, and diagnostics |
| Memory | Reuse facts, decisions, preferences, rules, and architecture across work |
| Task | Assign defined work to one or more connected Agents |
Normal Agent work can create a Session compilation and zero or more long-term memories. Create a Task from Dashboard -> Tasks, a Session detail, or a Memory detail when a connected Agent should execute work.
Advanced: Connect This Device
The TokST CLI connects local Agent endpoints, captures Sessions, executes Tasks, receives deliveries, and provides diagnostics. Cloud MCP remains the memory interface.
For macOS and Linux, copy and run this single command. It downloads the installer before execution:
curl -fsSLo tokst-install.sh https://tokst.com/install.sh && sh tokst-install.sh
tokst login
tokst start --agent all
tokst status
Windows PowerShell:
Invoke-WebRequest https://tokst.com/install.ps1 -OutFile "$env:TEMP\\tokst-install.ps1"
& "$env:TEMP\\tokst-install.ps1"
Use tokst start --agent <name> to connect one supported local Agent. The
Dashboard shows connected endpoints, active model, health, capabilities, and
recent Task activity.
Next Steps
- Read the MCP Server Reference for tool behavior.
- Read Memory Management for durable-memory quality rules.
- Use Agent Setup for local execution endpoints.
- Create and supervise formal work in Dashboard -> Tasks.