HREVN MCP Server
Minimal MCP server for HREVN, backed by the live managed runtime at https://api.hrevn.com.
Why this exists
For Anthropic and later Codex, MCP is the cleanest way to expose HREVN as real tools rather than only as local helper scripts. This server stays thin on purpose:
- canonical semantics remain in the private HREVN core
- execution remains in the live HREVN managed runtime
- the MCP layer only exposes HREVN capabilities in a standard tool interface
Included tools
baseline_checkprofile_validategenerate_bundleverify_bundle
Environment
export HREVN_API_BASE_URL="https://api.hrevn.com"
export HREVN_API_KEY="replace-me"
Install
cd hrevn-mcp-server
pip install -e .
If you install into a virtual environment, make sure the same virtual
environment is active when you launch Claude Code. Otherwise hrevn-mcp-server
may not be on the PATH Claude Code uses.
Run
hrevn-mcp-server
The server uses MCP stdio transport.
Verify before opening Claude Code
hrevn-mcp-server --version
hrevn-mcp-server --list-tools
hrevn-mcp-server --self-test
The self-test runs a live baseline_check against the configured HREVN managed API.
If hrevn-mcp-server is not found, use the absolute executable path from the
active environment, for example:
which hrevn-mcp-server
and place that path in your Claude Code MCP config instead of the bare command.
Claude Code MCP example
See:
docs/CLAUDE_CODE_MCP_CONFIG.mdexamples/baseline_check_payload.jsonserver.json
Recommended first test
Once Claude Code has this MCP server configured, ask it to call:
baseline_check
with the payload in:
examples/baseline_check_payload.json
That should return a real BaselineResult from the live HREVN managed API.
Design rule
This server must not reimplement HREVN truth locally. It should only expose stable MCP tools that call the managed API.
Registry readiness
This repo includes:
server.json
as a machine-readable manifest describing the server, transport, environment, and exposed tools for registry and catalog use.
Current status
This is a minimal first MCP server aimed at Anthropic-first testing. It is intentionally thin, real, and compatible with the live managed API path already in production.