Skip to main content
Glama
prismeai

Prisme.ai MCP Plugin

Official
by prismeai
README.md
# Prisme.ai MCP Plugin

Prisme.ai MCP is distributed as a plugin for **Claude Code** and **Codex**. The plugin bundles the MCP server, Prisme.ai skills, Claude agents, documentation, and the DSUL linter in one repository. Install and use the plugin only.

## What You Get

| Component | Description |
|-----------|-------------|
| MCP server | `prisme-ai-builder` tools for workspaces, automations, apps, events, files, AI Knowledge, and Prisme.ai documentation |
| DSUL validation | `validate_automation`, backed by the bundled linter |
| Skills | `/prisme-ai:*` skills for connector scaffolding, testing, documentation, fleet sync, A2UI, workspace pages, assistant workflows, and ticket validation |
| Claude agents | `code-review` and `prisme-assistant` for Claude Code |

## Install From GitHub

Repository: [prismeai/prismeai-mcp](https://github.com/prismeai/prismeai-mcp)

### Claude Code

In Claude Code:

```text
/plugin marketplace add prismeai/prismeai-mcp
/plugin install prisme-ai@prismeai-mcp
```

Then reload plugins or restart the session if the tools are not visible immediately.

### Codex

From a terminal:

```bash
codex plugin marketplace add prismeai/prismeai-mcp
codex plugin add prisme-ai@prismeai-mcp
```

The plugin source is `./plugin` inside this repository. Both marketplaces point there, so the same GitHub repo installs cleanly in Claude Code and Codex.

## Authenticate

Credentials are user-created API tokens, registered per environment. The recommended path keeps the token **out of the chat** (it is never sent to the LLM provider):

1. Create a token in the studio of the target environment: `https://<studio-domain>/settings/tokens` (e.g. <https://sandbox.prisme.ai/settings/tokens>).
2. Run the `set-token` command in your own terminal — the exact path + config dir are printed in the "no credentials" error:

   ```bash
   node "<plugin>/build/index.js" set-token sandbox --config-dir "<config-dir>"
   ```

   It prompts for the token with hidden input, probe-validates it against the API, then saves it to the plugin data dir (`credentials.json`, mode 600). An invalid token saves nothing.
3. Re-run your request — the server picks up the new token automatically (no restart). Run `set-token` again anytime to rotate.

When a tool call has no token (or hits a 401), the error message contains the exact command to run. You can instead let the agent register a pasted token via the `set_token` tool, but that token is sent to the LLM provider as part of the conversation — prefer the CLI.

## First Use

After installation, run:

```text
/prisme-ai:guide
```

The guide lists every bundled skill and includes the Prisme.ai environment rules, workspace parameter rules, event-search patterns, and recommended workflow.

For environment or token setup help, run:

```text
/prisme-ai:setup
```

Typical requests:

```text
List automations in ai-knowledge on sandbox
```

```text
Trace this correlationId in sandbox: <id>
```

```text
/prisme-ai:app-mcp-implement Salesforce connector
```

## Updating

Pull plugin updates from the marketplace:

### Claude Code

```text
/plugin marketplace update prismeai-mcp
```

### Codex

```bash
codex plugin marketplace upgrade prismeai-mcp
```

Release tags rebuild and commit the self-contained bundle at `plugin/build/index.js`.

## Runtime Model

The plugin starts the committed bundle:

```text
plugin/build/index.js
```

Runtime requirements:

- Node.js, provided by the host environment
- No `npm install`
- No local build
- No Playwright
- No browser token capture

## Maintainer Development

Only plugin maintainers need source-based local setup. Use [Development](./docs/DEVELOPMENT.md) to run an MCP client against this repository checkout and rebuild the committed runtime artifact.

## Plugin Layout

| Path | Purpose |
|------|---------|
| `.claude-plugin/marketplace.json` | Claude marketplace entry, pointing to `./plugin` |
| `.agents/plugins/marketplace.json` | Codex marketplace entry, pointing to `./plugin` |
| `plugin/.claude-plugin/plugin.json` | Claude plugin manifest |
| `plugin/.codex-plugin/plugin.json` | Codex plugin manifest |
| `plugin/.mcp.json` | Claude MCP server definition |
| `plugin/.codex-plugin/mcp.json` | Codex MCP server definition |
| `plugin/build/index.js` | Self-contained MCP server bundle |
| `plugin/skills/` | Bundled Prisme.ai skills |
| `plugin/agents/` | Claude Code agents |
| `plugin/llmDoc/` | Prisme.ai documentation exposed to tools |

## Reference Docs

| Guide | Description |
|-------|-------------|
| [Quick Start](./docs/QUICK_START.md) | Plugin install and first token setup |
| [Manual Setup](./docs/MANUAL_SETUP.md) | Configure non-plugin MCP clients against the committed plugin bundle |
| [Tools Reference](./docs/TOOLS.md) | MCP tools exposed by the plugin |
| [Environments](./docs/ENVIRONMENTS.md) | Plugin environment and token persistence |
| [Development](./docs/DEVELOPMENT.md) | Local repository setup for maintainers and release flow |

TDQS

A3.5/5.0

Scored across 39 tools

Disambiguation4/5

Most tools have distinct purposes, with clear naming distinguishing between operations like create, delete, get, list, update for different resources (automations, files, app instances). There is a slight potential for confusion between similar tools like get_app_instance and get_app_instance_config, but descriptions clarify. Overall, the set is well-organized and an agent can discriminate effectively.

Naming Consistency5/5

All tools follow a consistent verb_noun pattern using snake_case (e.g., create_automation, delete_file, get_app). No mixing of conventions. Even the legacy AI tools use a consistent 'ai_knowledge_' prefix followed by a noun. This high consistency aids predictability.

Tool Count2/5

With 39 tools, this server is over the typical well-scoped range. While each tool has a specific purpose, the number is overwhelming for an agent, especially given that some domains (e.g., app instances) have many specialized tools that could potentially be merged. This number would better fit a more focused scope.

Completeness4/5

The server covers CRUD for automations, files, app instances, and reports, plus workspace management, event search, documentation, and a generic API call tool. However, there is no explicit tool for deleting a workspace, which is a notable gap. The generic call_api can likely handle it, so it's a minor gap. Overall, the tool set is fairly complete for the platform's core operations.

Maintenance

ActivitySlowing
ResponsivenessSyncing