Heddle
A policy-and-trust layer that turns declarative YAML configs into Model Context Protocol tool servers with trust-tier enforcement, credential brokering, and tamper-evident audit logging built in.
Heddle turns a YAML configuration file into a complete MCP tool server with no Python or boilerplate required, running it with heddle run agents/config.yaml. Every tool call passes through trust enforcement, credential brokering, and audit logging: an agent assigned a read-only trust tier that attempts a POST request is blocked and logged, with the violation linked into a hash chain covering every event before and after it.
Credentials are resolved from a {{secret:key}} reference at runtime and never stored in the config itself, and each server can run inside a sandboxed Docker container with dropped capabilities, a read-only root filesystem, a process-count limit, and pinned image digests. A heddle generate command can produce a working config for wrapping an existing API from a natural-language description in about 20 seconds.
Multiple Heddle-defined tool servers can be meshed together behind a single MCP connection; a current demo environment runs 46 tools from 9 active configs this way.