# Policyseed MCP server

> Draft and manage SOC 2 governance policy documents through an AI assistant using a deterministic template engine.

- Repository: https://github.com/odedmoshe/policyseed
- Page: https://hostedmcp.org/servers/odedmoshe/policyseed
- Transport: stdio
- Stars: 1
- License: Apache-2.0
- Latest version: 1.1.1

## Install

### Claude Code

```sh
claude mcp add policyseed -- npx -y policyseed
```

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

```json
{
  "mcpServers": {
    "policyseed": {
      "command": "npx",
      "args": [
        "-y",
        "policyseed"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "policyseed": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "policyseed"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"policyseed","type":"stdio","command":"npx","args":["-y","policyseed"]}'
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "policyseed": {
      "command": "npx",
      "args": [
        "-y",
        "policyseed"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add policyseed -- npx -y policyseed
```

### Codex config.toml (`~/.codex/config.toml`)

```toml
[mcp_servers.policyseed]
command = "npx"
args = ["-y", "policyseed"]
```

### Gemini CLI

```sh
gemini mcp add -s user policyseed npx -- -y policyseed
```

### Gemini settings.json (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "policyseed": {
      "command": "npx",
      "args": [
        "-y",
        "policyseed"
      ]
    }
  }
}
```

## Tools (5)

- `list_policies`: List the 22 SOC 2 policy templates Policyseed can render: id, slug, title, one-line summary, owner role, output file name and the SOC 2 Trust Services Criteria each covers. Use the slug with render_policy. Every policy shares one review cadence, set by the intake's review_cadence field (Annual by default). (read-only)
- `intake_questions`: Return every intake field the policy renderer needs, with type, whether it is required, allowed values, length limits, default and a description. Call this first, then interview the user: ask for the required fields (company, product, security_owner, approver, incident_contact) and confirm or change the defaults (cloud, identity provider, data types, etc.) before calling render_policy or render_all. (read-only)
- `render_policy`: Render a single SOC 2 policy as Markdown from the company intake, exactly as `policyseed build` writes it (company-titled H1, 9 standard sections, revision history). Returns the Markdown text; suggest saving it as the file_name shown by list_policies. Invalid intake values are reported back so you can fix them. (read-only)
- `render_all`: Render all 22 SOC 2 policies (or the subset named in `slugs`) as Markdown from the company intake. Output is large (roughly 250 KB for all 22): the first content item is an index, then one item per policy starting with an HTML comment naming its file, e.g. <!-- file: 03-access-control-policy.md -->. Prefer render_policy when the user only needs one or two documents. (read-only)
- `check_reviews`: Run the same logic as `policyseed check`: read previously rendered policies (NN-<slug>.md files) and report which are overdue for their periodic review, based on the last date in each policy's revision history and the intake's review_cadence (Annual 365d, Semi-annual 182d, Quarterly 91d). Paths are on the machine running this server; pass `directory` as an absolute path to the project, since the server's working directory may not be the user's project. (read-only)

## FAQ

### Does Policyseed use an LLM to generate policies?

No, the policies are generated using plain Markdown templates with a Mustache-style tag syntax, ensuring deterministic output without model randomness.

### What kind of evidence can it collect for audits?

It can collect read-only evidence from AWS, Google Cloud, and GitHub using your existing authenticated CLIs.
