TokenSaver
EN

La documentation est disponible en anglais.

Documentation / Use cases

Agents & MCP

Coding agents and chat hosts (Cursor, Claude Code, Mistral Studio) call tools through MCP. TokenSaver’s Trust Gateway applies policy; the agentic graph shows the tool neighbourhood around each hub.

This is where TokenSaver’s innovation is most visible: not only “an LLM proxy”, but a trust layer for tools plus a cartography of what the agent actually did.

Agent loop
  1. IDE / chat host
  2. MCP client
  3. Tools / Trust Gateway
  4. TokenSaver audit
  5. Agentic graph
  1. Route the agent tokensaver route cursor / claude — or paste MCP URLs from doctor into the IDE config. For Mistral Studio, use OAuth against mcp.tokensaver.fr/mcp.
  2. Run a tool-using task Ask the agent to read a file or call a tokensaver_* tool so MCP traffic is generated.
  3. Open Flux IA then the graph Find the MCP / egress hub → « Voir dans le graphe agentique » → expand tools and open the inspector.
Agentic graph — hub + tool satellites
MCP Trust Gateway status (optional)

Related: Connect Cursor, Claude Code, Mistral Studio, Agentic graph, Build → MCP.