Three tools exposed
Only search, get, and execute reach the agent, not the full tool list.
MCP Tool Search
100+ tools, one route, a fraction of the context.
Expose every registered MCP server behind a single endpoint, without loading every tool definition into the model's context window just to make one call.
- 3
- tools exposed to the agent
- 100+
- tools behind one route
- Flat
- token cost as you add servers
The problem
Register enough MCP servers and the naive approach, listing every tool from every server up front, starts eating the context window before the agent has done anything. Most of those definitions are irrelevant to the task at hand, but the model still pays to read them, and accuracy drops as the tool list grows.
How it works
Instead of exposing the full tool catalog, Fabric exposes three: search, get, and execute. The agent searches for the tool it needs, retrieves only that tool's definition, and executes it. The other 99 definitions never enter the context window. Behind that thin interface, Fabric still fans out across every registered MCP server, internal or external, and every execution goes through the same credential brokering, RBAC, and audit logging as any other route.
What you get
On-demand definitions
Token cost stays flat as you add servers.
One route
100+ tools and MCP servers reachable behind a single route.
Swap backends freely
Add or swap backing servers without touching agent config.
Same policy layer
Auth, RBAC, and credential brokering as on LLM and API routes.
Bring every AI interaction under control
Connect your first model, MCP server or internal API and apply consistent governance from one gateway.