fix(mcp-utils): cap gateway tool name slugs to prevent 128-char limit

[viktormarinho]

What is this contribution about?

MCP clients like Claude Code prepend their own prefix to tool names (e.g. mcp__cms__). When combined with long connection ID slugs from GatewayClient.namespace(), the total tool name exceeded the 128-character limit enforced by Anthropic's API, causing 400 invalid_request_error for Claude Code users.

This adds a capSlug() function that truncates the slug portion of namespaced tool names to 32 characters. Proxied AI providers already went through sanitizeToolName() which truncates to 128 chars, but direct MCP clients (Claude Code, Codex) received raw untruncated names. Normal connection IDs (conn_<nanoid> ≈ 26 chars slugified) are unaffected.

How to Test

1. Create a Virtual MCP with connections that have long IDs
2. Connect Claude Code to the Virtual MCP endpoint
3. Verify tool calls succeed without 128-char API errors

Review Checklist

• PR title is clear and descriptive
• Changes are tested and working
• No breaking changes

---

Summary by cubic

Caps gateway tool name slugs to 32 characters in mcp-utils so full tool names stay under Anthropic’s 128-char limit. Fixes 400 errors in Claude Code and other MCP clients when connection IDs are long.

• Bug Fixes
  • Added capSlug() and MAX_SLUG_LENGTH = 32; applied when generating slugs in GatewayClient and when stripping namespaces.
  • Ensures consistent truncated slugs via an internal keyToSlug map; exports capSlug from the aggregate index.
  • Added tests for slug truncation, namespace stripping with long IDs, and correct tool routing with truncated slugs.

^{Written for commit 6160129. Summary will update on new commits.}

[github-actions]

🧪 Benchmark

Should we run the Virtual MCP strategy benchmark for this PR?

React with 👍 to run the benchmark.

Reaction	Action
👍	Run quick benchmark (10 & 128 tools)

Benchmark will run on the next push after you react.

[cubic-dev-ai]

No issues found across 3 files