MCP servers¶
The skills are instruction-only workflows and run without any MCP server. Three pinned servers make them sharper: two need no credentials, one needs an Anypoint login.
| Server and source | Pin | Role | Credentials |
|---|---|---|---|
mule-build |
@sfdxy/mule-build@2.2.0 |
Validation, testing, packaging, local runtime, versioning, security checks | None |
mule-lint |
@sfdxy/mule-lint@1.28.0 |
Canonical Mule standards, static analysis, and RAML/OAS validation | None |
anypoint-connect |
@sfdxy/anypoint-connect@0.12.0 |
Authorized Design Center, Exchange, Governance, runtime evidence, and mutations | Anypoint Platform login |
Each server has its own documentation, which is the place to look for command references, tool catalogs, and per-host setup beyond what the skills need:
| Server | Documentation |
|---|---|
mule-build |
https://avinava.github.io/mule-build/ |
mule-lint |
https://avinava.github.io/mule-lint/ |
anypoint-connect |
https://avinava.github.io/anypoint-connect/ |
Package links resolve to the exact registry version the checked-in configuration uses rather than an
unpinned latest release. Node.js >=20.19.0 satisfies all three.
Which skill uses which server¶
| Skill | mule-build |
mule-lint |
anypoint-connect |
|---|---|---|---|
mule-api-design |
No | AMF contract and local governance validation | Optional Design Center, Exchange, and centralized Governance |
mule-docs |
Inventory support | Optional | No |
mule-development |
Validation and tests after a change | Static analysis of changed files | No |
mule-testing |
Primary test execution and reports | Testing standards and guidance | No |
mule-troubleshooting |
Reproduction and validation | Optional | Telemetry for the incident window |
mule-ops |
No | No | Primary evidence source |
mule-review |
Validation during review | Static analysis | Optional runtime verification only |
mule-build |
Primary | Static and security gate | Only for authorized publish or deploy |
Nothing in this table is required. A missing server becomes a disclosed coverage gap, not a failed workflow.
Capabilities by server¶
mule-build — project readiness and resolved configuration, build and validation runs, local
runtime start, stop, and status, security enforcement and secure-property handling, and version and
release operations.
mule-lint — full-project lint analysis, AMF-backed RAML/OAS validation and local governance
profiles, rule detail lookup, single-snippet validation, and Mule XML formatting.
anypoint-connect — identity and environment discovery; application status, deployment
specification, resources, and settings; log retrieval, error analysis, log patterns, and log
statistics; performance, worker, memory, and time-series metrics plus AMQL queries; lifecycle
operations such as restart, scale, deploy, rollback, stop, start, and delete; Exchange search and
publication; API-manager instances, policies, and alerts; preview-bound Design Center project creation,
file sync, Exchange publication, and Governance reads; audit log;
Anypoint MQ queues and dead-letter inspection; and Object Store keys and values.
Lifecycle and mutating operations are never part of establishing readiness, and the skills require explicit authorization before any of them.
Configuration per host¶
| Host | File | Shipped form |
|---|---|---|
| Claude Code | bundled in the plugin | .mcp.json in the plugin |
| Codex | .codex/config.toml |
install/hosts/codex/config.toml |
| VS Code and Copilot in VS Code | .vscode/mcp.json |
install/hosts/vscode/mcp.json |
Copilot CLI, Gemini, other hosts accepting mcpServers |
.mcp.json |
install/hosts/mcp.json |
The generic form:
{
"mcpServers": {
"anypoint-connect": {
"command": "npx",
"args": ["-y", "@sfdxy/anypoint-connect@0.12.0", "mcp"]
},
"mule-build": {
"command": "npx",
"args": ["-y", "@sfdxy/mule-build@2.2.0", "mcp"]
},
"mule-lint": {
"command": "npx",
"args": ["-y", "@sfdxy/mule-lint@1.28.0", "mcp"]
}
}
}
The installer merges only the entries your configuration lacks and never overwrites an existing server of the same name. See Install for other agents.
Local MCP files do not configure GitHub-hosted Copilot agents or code review; configure hosted MCP access through repository settings. Claude Code asks for approval before using project-scoped MCP servers.
Verify and disable¶
| Host | Verify | Disable one server |
|---|---|---|
| Claude Code | /mcp |
/mcp, then disable it |
| Codex | codex mcp list |
Remove its [mcp_servers.<name>] table |
| Copilot CLI | copilot mcp list |
Remove its entry from .mcp.json |
| VS Code | Reload the window and inspect the MCP server list | Remove its entry from .vscode/mcp.json |
| Gemini and others | The host's MCP status view | Remove its entry from the host's config |
The first session after installing pays one npx cold start per server while the packages download.
Later sessions use the cache.
Keeping the pins consistent¶
The pinned versions appear in the plugin .mcp.json, the three host forms, the installer, and this
documentation. tools/validate_repository.py fails the build when any of them disagree, so a
version bump has to be applied everywhere or not at all.