Skip to main content
Glama
darthzen

jev-mcp

by darthzen
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Hybrid

jev-mcp

MCP server for TypeSafe Jev (System One): one tool, evaluate, whose input is the exact body of POST /v1/systemone and whose output is the exact response. Streamable HTTP transport, built on fastmcp 3.

Why one tool: every Jev client in this lab (Claude Code's jev.py, the Hermes skill, this server) then asks Jev with the same contract, so questions and results move between them unchanged.

Configuration

Variable

Meaning

Default

JEV_API_KEY / JEV_API_KEY_FILE

TypeSafe key (file wins; contents stripped)

required

JEV_MCP_PATH_SECRET / ..._FILE

random path segment; endpoint becomes /<secret>/mcp

none → /mcp with a warning

JEV_API_URL

TypeSafe endpoint

https://api.typesafe.ai/v1/systemone

JEV_MODEL

default model

jev-latest

JEV_BIND_HOST / JEV_BIND_PORT

bind address (not JEV_MCP_PORT: Kubernetes injects that name for a Service called jev-mcp)

0.0.0.0 / 8080

JEV_MCP_TIMEOUT

seconds per TypeSafe call

60

GET /healthz returns ok for probes. Any path other than the MCP path and /healthz is 404.

The server does not authenticate callers. When it is reachable from the internet the path secret is the credential: whoever has the full URL can spend the key's Jev credits (and nothing else, the process holds no other access). Treat the URL like a key.

Related MCP server: Jev MCP Proxy

Run locally

uv venv && uv pip install -e '.[dev]'
JEV_API_KEY_FILE=~/Developer/keys/jev/claude.key JEV_MCP_PATH_SECRET=$(openssl rand -hex 24) .venv/bin/jev-mcp
.venv/bin/pytest

Image

Built in-cluster with the lab-image-build skill (kaniko → Harbor), base registry.suse.com/bci/python:3.13, runs as uid 10001:

~/.claude/skills/lab-image-build/scripts/kaniko-build.sh \
  --repo darthzen/jev-mcp --image jev-mcp --tag 0.1.2

Deployment lives in lab-fleet/09-mcp/jev/ (Deployment, Service, Ingress on jev-mcp.ash4d.com, exposed through the Cloudflare tunnel).

Clients

# Claude Code
claude mcp add --transport http --scope user jev https://jev-mcp.ash4d.com/<secret>/mcp
# claude.ai / Claude Desktop: Settings → Connectors → Add custom connector → that URL, no OAuth
# Claude Desktop fallback if the connector flow insists on OAuth:
"jev": { "command": "npx", "args": ["-y", "mcp-remote", "https://jev-mcp.ash4d.com/<secret>/mcp"] }

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • jev-mcp

    arunav25
    A
    license
    A
    quality
    C
    maintenance
    Enables MCP clients to call TypeSafe's JEV classifier and receive structured, typed judgments with probabilities for binary, choice, and scoring questions.
    1
    7
    MIT

  • Jev MCP Proxy

    altregubov
    A
    license
    B
    quality
    C
    maintenance
    Enables AI agent skills to route state-evaluation requests to TypeSafe AI's Jev System One model, supporting typed questions, choice classification, binary probabilities, and rubric scoring with calibrated confidence.
    4
    MIT

  • mcp_jev

    pedroknigge
    A
    license
    A
    quality
    B
    maintenance
    Runs TypeSafe Jev System One packs locally, enabling agents to perform typed Choice, Noul, and Score judgments for tasks like PR auditing, intent routing, and locale classification.
    5
    1
    MIT