Remote MCP Server

Connect Cromus to your AI client

Cromus is a remote MCP server. Nothing to download or install: connect once, then score and validate skills without leaving your AI tool.

https://mcp.cromus.ai/mcp

What you get

score_skillReturns a 0 to 1000 Croms score with a four-dimension breakdown for any skill file.
simulate_costEstimates per-run and monthly cost across five routing modes for a given workflow.
score_graphImports GRAPH.md or a supported agent framework graph and returns graph-aware Croms findings.
score_harnessScores Cromus STACK.md, harness.md, or .harness/ source bundles with canonical engines.
simulate_graph_costEstimates bounded entry-to-exit path cost, latency, and failure risk for an agent graph.
check_mcp_driftCompares a skill's declared MCP dependencies against a live server and flags schema changes, missing tools, and new tools.
check_model_driftDetects when a skill's declared model has been retired, repriced, or superseded and suggests a validated replacement.
validate_skillChecks a SKILL.md file against the open specification and lists any violations.
validate_mcp_depsChecks a skill's mcp_dependencies block for completeness, valid transports, auth declarations, and fallback behavior.
validate_ethosChecks an ETHOS.md governance file against the open specification.
validate_harnessValidates Cromus STACK.md, harness.md, or .harness/ bundles and reports their detected format.
validate_memoryChecks a MEMORY.md handoff file against the open specification.
validate_agentsChecks an AI department configuration (Specialist, Manager, CWO roles) for completeness, scope conflicts, and governance coverage.
reality_check_summaryReturns the most recent Reality Check record for a skill, read-only.
check_usage_wasteAnalyzes AI API usage exports for waste patterns: low cache utilization, verbose output ratios, premium model concentration, spend acceleration, model sprawl, and idle burn. Accepts raw CSV or pre-aggregated monthly rows.
check_policyEvaluates a SKILL.md against a cromus.policy.yml governance policy (min_croms, max_cost_per_run, require_ethos, allowed_providers, and more). Returns pass/fail with per-rule violation details for CI and agent workflows.
get_approved_skillRetrieves an explicitly approved, versioned SKILL.md revision with its SHA-256 hash and canonical Git lineage. Read-only; requires an Ops+ Sync Center workspace.

Connect your client

Claude
Claude Code
Agent Plugin
DeepSeek Harness
VS Code
GitHub Copilot
Other MCP clients

Path A: From the directory

  1. 1Open Claude and go to Settings, then Connectors.
  2. 2Search for Cromus in the connector directory.
  3. 3Click Connect on the Cromus listing.
  4. 4Approve access on the Cromus consent screen.

Path B: Custom connector

  1. 1Open Claude and go to Settings, then Connectors.
  2. 2Choose Add custom connector.
  3. 3Paste https://mcp.cromus.ai/mcp into the server URL field.
  4. 4Click Add.
  5. 5Approve access on the Cromus consent screen.
https://mcp.cromus.ai/mcp

Available on Claude Pro, Max, and Team plans.

Claude

Claude Code

Agent Plugin

DeepSeek Harness

VS Code

GitHub Copilot

Other MCP clients

Validate a skill from inside Claude

Once the connector is active, you can score and validate any skill without leaving the conversation.

  1. 1

    Create or record a skill in Claude.

    Build a skill during your conversation, or describe a workflow you want to evaluate.

  2. 2

    Ask Claude to validate it.

    “Score the skill you just created using Cromus, validate its MCP dependencies, check my usage data for waste patterns, and summarize the findings.”

  3. 3

    Read the results.

    The Croms score, cost simulation, and structural findings appear in the conversation.

Claude reads the skill. Cromus scores it. You decide what ships.

Troubleshooting

The consent screen shows an error

Cromus does not appear in my client's tool list

Authorization window never opens

Which plans support connectors?

I see fewer tools than expected in my client

Prefer the web app? Everything the connector does is also available at cromus.ai.

Claude

Path A: From the directory

  1. 1Open Claude and go to Settings, then Connectors.
  2. 2Search for Cromus in the connector directory.
  3. 3Click Connect on the Cromus listing.
  4. 4Approve access on the Cromus consent screen.

Path B: Custom connector

  1. 1Open Claude and go to Settings, then Connectors.
  2. 2Choose Add custom connector.
  3. 3Paste https://mcp.cromus.ai/mcp into the server URL field.
  4. 4Click Add.
  5. 5Approve access on the Cromus consent screen.
https://mcp.cromus.ai/mcp

Available on Claude Pro, Max, and Team plans.

Claude Code

  1. 1Run the command below in your terminal.
  2. 2A browser window opens asking you to approve access.
  3. 3Approve access on the Cromus consent screen.
  4. 4Return to the terminal. The connector is active.
claude mcp add --transport http cromus https://mcp.cromus.ai/mcp

Agent Plugin

Install the Cromus Agent Plugin to get all 17 Cromus tools in any compatible coding agent — Claude Code, Cursor, Cline, Codex — whether or not you route through Vercel AI Gateway.

  1. 1Run the command below in your terminal inside your project.
  2. 2A browser window opens asking you to approve access.
  3. 3Approve access on the Cromus consent screen.
  4. 4All 17 Cromus tools are now available in your agent session.
npx plugins add @cromus-ai/plugin

Requires an agent client that supports Agent Plugins 1.0. The plugin is published as @cromus-ai/plugin on npm — no server to host or keys to manage.

DeepSeek Harness

Install the Cromus dsh-plugin to get all 17 Cromus tools in your DeepSeek Harness agent session.

  1. 1Run the install command below in your terminal.
  2. 2Set your Cromus API key as an environment variable (get one at cromus.ai/connect).
  3. 3Start or restart your dsh agent session.
  4. 4All 17 Cromus tools are now available under the mcp__cromus__ namespace.
dsh plugins add @cromus-ai/dsh-plugin
export CROMUS_API_KEY=your_key_here

Tools appear under the mcp__cromus__* namespace. The plugin is published as @cromus-ai/dsh-plugin on npm — no server to host or keys to manage beyond the API key above.

VS Code

The Cromus VS Code extension scores and validates your governance files directly in the editor — no context switching required. Inline diagnostics, status-bar CROMS score, model pricing on hover, and policy-as-code enforcement all run on save.

  1. 1Search for Cromus in the VS Code Marketplace and click Install, or run the command below.
  2. 2Open VS Code Settings (Cmd/Ctrl+,) and search for Cromus.
  3. 3Set Cromus: Api Key to your Cromus API key (get one under API Keys above).
  4. 4Open any SKILL.md — your CROMS score appears in the status bar and inline diagnostics run on save.
code --install-extension CromusAI.cromus
SettingDefaultDescription
cromus.apiKey""Your API key (cromus_ci_… or mcp_…)
cromus.apiUrlhttps://cromus.aiOverride for self-hosted instances
cromus.validateOnSavetrueRun validation automatically on save
cromus.showScoreInStatusBartrueShow CROMS score in the status bar

View on the VS Code Marketplace. The same API key works in the MCP server, the GitHub Action, and the extension.

GitHub Copilot

The Cromus GitHub Copilot Extension puts scoring, validation, and cost simulation directly inside Copilot Chat — on GitHub.com, in VS Code Copilot Chat, and anywhere Copilot Chat extensions are supported. No API key or MCP server needed.

  1. 1Click the button below to install the Cromus GitHub App on your account or organization.
  2. 2Open any GitHub repository and click the Copilot Chat icon.
  3. 3Type @cromus followed by your request — no API key or MCP server required.
  4. 4Cromus responds inline: scores, validation results, cost tables, and policy checks.
CommandWhat it does
@cromus score [paste SKILL.md]Returns a Croms score with four-dimension breakdown and recommendations
@cromus validate [paste SKILL.md]Checks against the open SKILL.md specification
@cromus score harness [paste STACK.md or harness bundle]Scores canonical components, Structural Gaps, and Graph Layer signals across supported formats
@cromus validate harness [paste STACK.md or harness bundle]Validates Cromus STACK.md, harness.md, or .harness/ source bundles deterministically
@cromus simulate [paste SKILL.md]Estimates cost across all five routing modes
@cromus check policyRuns policy-as-code enforcement — paste your skill and a cromus.policy.yml block
@cromus helpLists all available commands

All analysis is deterministic — zero additional LLM calls. Paste your SKILL.md content directly into the chat; fenced code blocks and bare YAML both work.

Other MCP clients

Any client that supports remote MCP servers with OAuth can connect using the server URL below. Add the following block to your client's MCP server configuration. The client handles the OAuth authorization flow automatically.

{
  "mcpServers": {
    "cromus": {
      "type": "http",
      "url": "https://mcp.cromus.ai/mcp"
    }
  }
}

After saving the configuration, start a new session. The client will open a browser window for authorization on the first connection.

The consent screen shows an error

Try again in a fresh private browsing window. If the error persists, contact support at cromus.ai/contact.

Cromus does not appear in my client's tool list

Disconnect and reconnect the connector, then start a new conversation.

Authorization window never opens

Check that pop-ups are allowed for your AI client.

Which plans support connectors?

Connectors require a paid plan on most clients. Cromus itself does not charge for connecting.

I see fewer tools than expected in my client

Disconnect and reconnect the Cromus connector, then start a new conversation. Clients cache the tool list at connection time.