Jevbridge MCP
An MCP and ACP adapter for the TypeSafe Jev decision model: typed answers and confidence gates for clicks and tool calls.
- 51 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The adapter sends state to an external model and decides whether a click or tool call runs.
- API keys are set through environment variables.
Install
- Works in
Manual install
git clone https://github.com/tacticocc/Jevbridge.gitThe first install step from the README: clone, then `npm install`. The `@tactico/jevbridge` package arrives after a release is published.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The adapter sends state to an external model and decides whether a click or tool call runs.
- API keys are set through environment variables.
This is third-party code. Review the repository files before installing.
What it does
Jevbridge runs alongside the model you already use. TypeSafe Jev returns typed decisions rather than text, so it suits routing, scoring and action selection. With `TYPESAFE_API_KEY` set the adapter uses Jev directly, otherwise it takes Codex, Claude, Grok, OpenCode or any OpenAI-compatible API and makes it answer in the same format. The adapter evaluates one state against mixed questions of three kinds: choice, score and Noul. For tool calls and interface clicks there are confidence gates with outcomes `execute`, `confirm`, `escalate` and `abort`. The MCP tools are `jev_decide`, `jev_gate` and `jev_computer_use`, and ACP over stdio lets Zed and JetBrains keep Jev as a decision sidecar. Recipes cover support routing, dangerous command gates and picking the next interface action.
Who it is for. Agent developers who need controlled action selection and confirmation before a risky step.
Good fit when
- An agent is about to click a delete button and needs a confidence gate before executing.
- Support tickets must be routed by a typed model answer.
- Tests and CI need an offline heuristic mode with no network.
- An ACP-capable editor such as Zed should get Jev as a decision sidecar.
Not a fit when
- You need text generation or chat with no typed decisions.
- State cannot be sent to an external LLM.
- You need a ready npm package with no repository clone.
Example request
Use jev_gate to decide whether to run the click on the delete account button or ask for confirmation first.MCP
- Transport
- stdio
- Authentication
- key is optional
Environment variables
TYPESAFE_API_KEYsecret- TypeSafe key for native Jev. Without it the adapter uses heuristics or your LLM.
JEVBRIDGE_LLM- LLM provider choice, for example xai.
XAI_API_KEYsecret- Key of the chosen LLM provider. OPENAI_API_KEY, ANTHROPIC_API_KEY and OPENCODE_API_KEY also work.
Limitations
Node.js 22 or newer and Git are required. A TypeSafe key is optional: without it the heuristic mode and any OpenAI-compatible LLM work, but then an ordinary model makes the decisions instead of Jev. The `@tactico/jevbridge` package is named as available after a release is published, and until then the adapter runs from a clone with `node --experimental-strip-types`. Confidence gates reduce the risk of a wrong click but do not replace human oversight, especially for irreversible actions.
How to disable. Remove the `jevbridge` entry from the client MCP configuration and delete the repository clone. Remove the keys from environment variables.
FAQ
How do I connect it to Cursor?
Add a `jevbridge` entry to the MCP configuration with the command `node`, arguments `--experimental-strip-types`, the path to `src/cli.ts` and `mcp`, as in `examples/cursor.mcp.json`.
Is a TypeSafe key required?
No. Without `TYPESAFE_API_KEY` the adapter uses a heuristic mode or your OpenAI-compatible LLM.
Which outcomes do confidence gates give?
`execute` for a confident answer, `confirm` for middling confidence, plus `escalate` and `abort` for doubtful cases.
Which MCP tools are available?
`jev_decide`, `jev_gate` and `jev_computer_use`.
How can I try it offline?
Run `node --experimental-strip-types src/cli.ts eval support-route`: the recipe runs in heuristic mode with no network.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server giving an agent three Jev tools: typed yes or no, choice and score questions with probabilities, offline by default.
A plugin and MCP server that hand no-text decisions to the Jev model and gate shell commands while the LLM writes content
A plugin with a catalogue of public Jev examples: a skill and an MCP server show how others solved the same decision task.
Jev-Style runs small decision models on your machine: a probability per option, six skills, MCP and a Claude Code guard hook
The Jev classifier as an agent tool: labels items, answers yes or no and ranks candidates with probabilities within milliseconds.
MCP server that gives an agent probabilities and a pick from set options, like ticket urgency, so code can branch on a number
README in short
The adapter combines state evaluation, confidence gates for risky steps and two connection points: MCP and ACP over stdio. It runs on Node.js with no runtime dependencies.