One line of config, nine tools: what should I watch out for?

Disabling a tool hides it from discovery and denies it at call time. Clients cache discovery, so restart after a policy change.

One line of config, nine tools: what should I watch out for?

Remote MCP needs a bearer key; local stdio uses the CLI's stored credential.

INTEGRATION

One line of config, nine tools

The same core, exposed as MCP tools and resources. Works in Claude, Cursor, and any MCP client, with per-workspace tool policies.

CLIENT → SERVER → GRAPH

01ConnectPoint the client at the SSE URL with a bearer key.

02DiscoverTools and resources appear with their schemas.

03PolicyPer-workspace toggles hide tools you do not want exposed.

04AuditEvery tool call lands in the audit log with its actor.

Tool visibility is a workspace policy, so a client only ever sees what that workspace allows.

Calls

picx.generate_image: Text or reference image in, finished still out.
picx.edit_chain: Run a chain of edit operations on any asset.
picx.get_job: Poll async job status and collect the result URL.
picx.list_assets: Browse everything the workspace has generated.

Requestclaude_desktop_config.json

{
  "mcpServers": {
    "picx": {
      "url": "https://mcp.picxstudio.com/sse"
    }
  }
}

Response200 OK

✓ picx connected
✓ 9 tools · 4 resources

From the CLI

$ npx @picx/mcp install --client claude

✓ config written
✓ restart Claude to load 9 tools

For users

One line in your config and your assistant can shoot. No plugin, no desktop bridge, no local server.

For developers

Tools are coarse on purpose. An agent should reach a first cut in a dozen calls, not two hundred — fine-grained tools burn context and confuse planning.

Gotchas

Disabling a tool hides it from discovery and denies it at call time. Clients cache discovery, so restart after a policy change.
Remote MCP needs a bearer key; local stdio uses the CLI's stored credential.