Jambonz setup mcp

Setting Up the jambonz MCP Server

The jambonz MCP server lets AI coding tools query jambonz schemas, SDK examples, and the developer toolkit on demand — no manual docs hunting, no copy-pasting verb definitions.

What the Server Provides

Three tools, exposed via the Model Context Protocol:

• jambonz_developer_toolkit — The full developer guide. Verb model, transport modes, callback patterns, schema index. Load first when starting any jambonz work.
• get_jambonz_schema — On-demand JSON Schema for any verb, component, callback, or guide (e.g. verb:say, component:synthesizer, callback:actionHook, guide:node-sdk).
• get_sdk_example — Source code for any @jambonz/sdk example (agent, ivr-menu, dial, call-recording, etc.).

Two Transports

Transport	When to use	Endpoint
HTTP (remote)	Default for most users. No install. Stays current automatically.	https://mcp-server.jambonz.app/mcp
stdio (local)	When the IDE requires stdio, or when you want to run offline / behind a firewall.	npx -y @jambonz/mcp-schema-server

Setup by IDE

Claude Code

One command:

claude mcp add --transport http jambonz https://mcp-server.jambonz.app/mcp

Or for stdio:

claude mcp add jambonz -- npx -y @jambonz/mcp-schema-server

Confirm with claude mcp list — jambonz should appear with status connected.

Cursor

Edit ~/.cursor/mcp.json (global) or <project>/.cursor/mcp.json (project-scoped):

{
  "mcpServers": {
    "jambonz": {
      "url": "https://mcp-server.jambonz.app/mcp"
    }
  }
}

For stdio:

{
  "mcpServers": {
    "jambonz": {
      "command": "npx",
      "args": ["-y", "@jambonz/mcp-schema-server"]
    }
  }
}

Reload Cursor; the server appears under Settings → MCP.

VS Code (with Copilot or Continue / other MCP-aware extensions)

Add to .vscode/mcp.json in the project (or the global equivalent your extension reads):

{
  "servers": {
    "jambonz": {
      "type": "http",
      "url": "https://mcp-server.jambonz.app/mcp"
    }
  }
}

For stdio, use "type": "stdio" with "command" and "args" as above.

Codex

Codex reads ~/.codex/config.toml. Add:

[mcp_servers.jambonz]
command = "npx"
args = ["-y", "@jambonz/mcp-schema-server"]

(Codex currently uses stdio MCP. If/when HTTP support lands, switch to the URL form.)

Windsurf

Edit ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "jambonz": {
      "serverUrl": "https://mcp-server.jambonz.app/mcp"
    }
  }
}

Restart Windsurf.

Any other MCP-aware tool

For HTTP transport, point at https://mcp-server.jambonz.app/mcp. For stdio, run npx -y @jambonz/mcp-schema-server.

Verification

After setup, three tools should be available:

• jambonz_developer_toolkit
• get_jambonz_schema
• get_sdk_example

Quickest check: prompt the assistant with "Use the jambonz_developer_toolkit tool and summarize what it returns." If you get a verb list and schema index, the connection is working.

Pairing with the jambonz Skills

The MCP server and the jambonz-skills plugin complement each other — install both.

Skills tell the model what to do and why. The MCP server tells the model the exact JSON shape to write. Skills capture judgment (decision trees, patterns, gotchas) and are loaded into the model's context. The MCP server captures truth (live schemas, SDK example source) and is called as tools at runtime.

You can use them separately — skills alone for planning/offline work, MCP alone for one-off schema lookups — but real development wants both: skills decide, MCP verifies.

Install the skills with:

npx skills add jambonz/skills

Or copy the skills/ directory of the jambonz/skills repo into your agent's skills path manually.

Troubleshooting

Symptom	Likely cause	Fix
Tools don't appear after config change	IDE didn't reload	Restart the IDE (or its MCP subsystem)
connection refused on HTTP	Network / proxy issue	Try stdio transport (npx -y @jambonz/mcp-schema-server)
command not found: npx on stdio	Node not installed	Install Node.js ≥ 18
Tools appear but always error	Stale package version	Pin to latest: npx -y @jambonz/mcp-schema-server@latest
AI doesn't call the tools	Description not triggering	Explicitly prompt: "Call get_jambonz_schema('verb:dial') and show the result."