# evalship

> evalship is a freemium MCP server for developer tools. Hosted MCP server that lets coding agents read eval to-dos, production issues and PR. It runs as a hosted remote server, so there is nothing to install locally.

- **Kind:** MCP server
- **Pricing:** Freemium
- **Starting price:** $29 a month
- **Platforms:** API, CLI, Web
- **Website:** https://evalship.com/docs/agent
- **MCP Registry name:** com.evalship/evalship
- **Categories:** [developer tools](https://ai-powered.com/categories/developer-tools)
- **Listing updated:** 2026-11-18

## About

The README-style registry entry describes a server that reads a repository's to-do list, production issues and pull request reports from evalship. The server exposes four read tools: todo, issue, calls and pull_request. Access uses a read token that the user approves in a browser, and the server changes nothing by default. An optional flag adds a fifth tool that can propose a dismissal of a flagged flaw, which still needs someone with push access to accept it.

The server sits on top of evalship's hosted eval service, which runs LLM evals from GitHub Actions and posts results on pull requests. It is aimed at developers who already use evalship and want their coding agent to pull failing-eval context into its work. The docs also say production calls are visible only to repository members, even on public repositories.

The site describes a paid hosted plan alongside a free tier, but the MCP server itself is reached through the evalship account rather than installed locally.

## Features

- **Read-only repository tools:** Four read tools return open to-do items, production issues, AI call records and pull request eval reports.
- **Browser-approved read token:** Setup creates a read token through a browser approval page before adding the server to a client.
- **Optional dismissal proposals:** An opt-in fifth tool lets the agent propose dismissing a flaw, which a person must accept on the report.
- **Remote streamable HTTP endpoint:** The server is served over streamable HTTP at a single URL with a bearer token header.

## Good for

- Letting a coding agent check which eval flaws block a pull request before editing code
- Pulling recent failing production AI calls into a debugging session
- Reading open eval to-do items for a repository from inside an agent client

## Install

### Claude Code (Terminal)

```sh
claude mcp add --transport http evalship https://evalship.com/mcp
```

### Claude Desktop (Settings → Connectors → Add custom connector)

```sh
https://evalship.com/mcp
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "evalship": {
      "url": "https://evalship.com/mcp"
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "evalship": {
      "type": "http",
      "url": "https://evalship.com/mcp"
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "evalship": {
      "serverUrl": "https://evalship.com/mcp"
    }
  }
}
```

### Grok Bot (Grok Bot desktop app → Connect apps → Add a custom connector)

```sh
https://evalship.com/mcp
```

Enter a Name (for example "evalship"), choose Type: Remote HTTPS, paste this as the Server URL and choose Add plugin. When we checked, this server asked for a bearer token. Add it under HTTP headers as Authorization: Bearer <your token>.

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "evalship": {
      "type": "http",
      "url": "https://evalship.com/mcp"
    }
  }
}
```

## Tools

- `todo`: Returns the repository's open To do items with links and fix prompts.
- `issue`: Returns one production issue with its failing calls, latest trace and suggested fix.
- `calls`: Lists production AI calls, newest first, with optional filters.
- `pull_request`: Returns a pull request's eval report: which evals broke, why, and the flaws.

## FAQ

### Can the server change anything in my repository?

By default no. The docs say a read token reads what you see on evalship and changes nothing, and a proposal dismisses no flaw until someone accepts it.

### Which tools does the server expose?

Four read tools are listed: todo, issue, calls and pull_request. A fifth, propose_dismissal, appears only if the user opts in during setup.

### Does it work with chat clients like claude.ai?

Not yet. The docs say claude.ai and other chat clients that sign in with OAuth are not supported.

### Is evalship free?

evalship has a free plan, with paid plans for more usage or features. Paid plans start at $29 a month.

### How do I add evalship to Claude Code?

Run this command in your terminal: claude mcp add --transport http evalship https://evalship.com/mcp

### What are the best alternatives to evalship?

The closest alternatives to evalship are EvalGate, Errilo, Feedbacks, PathToShip: production readiness for AI-built apps and CueApril.

## Where this comes from

- Description: written from the project's website and its MCP Registry entry.
- Status: checked automatically by AIPoweredBot on 19 November 2026.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [EvalGate](https://ai-powered.com/mcp/evalgate): EvalGate is a freemium MCP server for ai models and llm tooling.
- [Errilo](https://ai-powered.com/mcp/errilo): Errilo is a freemium MCP server for developer tools.
- [Feedbacks](https://ai-powered.com/mcp/feedbacks): Feedbacks is an open-source MCP server for developer tools.
- [PathToShip: production readiness for AI-built apps](https://ai-powered.com/mcp/pathtoship-production-readiness-for-ai-built-apps): PathToShip: production readiness for AI-built apps is a free MCP server for developer tools.
- [CueApril](https://ai-powered.com/mcp/cueapril): CueApril is an MCP server for developer tools.
- [EvalGate Docs](https://ai-powered.com/mcp/evalgate-docs): EvalGate Docs is a freemium MCP server for ai models and llm tooling.
- [memi](https://ai-powered.com/mcp/memi-2): memi is an open-source MCP server for developer tools.
- [repo-interview](https://ai-powered.com/mcp/repo-interview): repo-interview is an open-source MCP server for developer tools.

---
Source: https://ai-powered.com/mcp/evalship
