Toolport
A local MCP gateway for sharing MCP servers across multiple AI clients after one setup, with keys kept on your machine.
About Toolport
Toolport is a local MCP gateway that lets you configure an MCP server once and share it across AI clients such as Claude, Cursor, and Codex. Tools load on demand to reduce context use, while keys stay on your machine. The official page says cross-server tool-definition tokens can fall 74% to 91% at the same task success rate. It fingerprints each tool to flag rug pull and tool poisoning; checks are enabled by default and run fully locally. API keys stay in the operating system keychain and are injected at runtime. It also imports CLAUDE.md or AGENTS.md agent rules for use across clients, with no account or cloud dependency.
Best for
Best use cases
Configure an MCP server once and reuse it across multiple AI clients
Use on-demand tool loading to reduce context and token use
Detect changed tool definitions or hidden instructions in descriptions
Configure a shared, governed MCP server set for a team
Features
Multi-client MCP server sharing
Configure an MCP server once and share it across AI clients such as Claude, Cursor, and Codex; keys stay on your machine.
On-demand tool loading
Tool definitions load on demand to control context use; the official page reports a 74% to 91% reduction in cross-server tool-definition tokens at the same task success rate.
Tool integrity checks
Fingerprints each tool to flag rug pull (a definition changed after approval) and tool poisoning (hidden instructions in a description); enabled by default and fully local.
Keychain and agent rules
API keys are stored in the operating system keychain and injected at runtime; import CLAUDE.md or AGENTS.md rules and apply them to clients such as Claude Code, Codex, Gemini CLI, and Zed.
Pros
- Open source (MIT license), with fully reviewable code
- Tool integrity checks enabled by default and run entirely locally
- API keys stored in the operating system keychain and injected at runtime
- Official benchmark reports a 74% to 91% reduction in cross-server tool-definition tokens at the same task success rate