gz-mcp-lsp
Wraps a single Astral ty language server (LSP over stdio) and exposes its type-aware Python semantic navigation over MCP: go-to-definition, references, diagnostics, hover with semantic type information, document and workspace symbols, implementations, and rename_symbol/edit_file with operator-enabled preview/apply, workspace containment and conflict checks.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
gz-mcp-lsp
An independent Python counterpart to Isaac Phi’s mcp-language-server, focused on reliable ty-backed semantic navigation for Codex through MCP.
Status: plan approved on 2026-11-01; implementation and native host evidence complete through T11; paused at a stage boundary. All nine tools have installed MCP/ty evidence and actual Codex invocation/approval checks; the full milestone remains unaccepted.
The intended path is:
Codex → MCP (stdio) → gz-mcp-lsp → LSP (stdio) → ty serverThe proposed scope is one workspace and one Astral ty process. Functional parity includes all six reference tools: definition, references, diagnostics, hover, rename_symbol and edit_file. Exact URI/identifier positions correct ambiguous name-resolution behavior. New tools are limited to document symbols, workspace symbols and implementations; hover carries semantic type information.
Rename/edit use operator-enabled preview/apply with workspace containment, conflict checks and durable recovery; the design does not promise portable atomic multi-file changes. No request-time installation, network service, second language server or general-purpose LSP framework is planned. The complete milestone is accepted parity, semantic corrections/extensions, real Codex interoperability, engineering evidence, then a deliberate stop.
Continue from the bounded implementation handoff; T12 is next after the final exact-commit platform gate. See T11 acceptance evidence for verification limits.
Read the revised architectural specification, version-pinned parity assessment, acceptance matrix, and approved implementation plan. Current design-revision evidence is separate from the historical initialization evidence. See the license review for attribution/dependency boundaries.
Development
Use uv 0.13.6 and its managed CPython 3.14.10. The exact interpreter is in
.python-version; dependencies and tools are resolved in uv.lock. This is a
standalone src/gz_mcp_lsp package using uv_build. Version 0.1.0 identifies the
unreleased development package, independent of starter version 0.7.1.
git clone https://github.com/tvproductions/gz-mcp-lsp.git
cd gz-mcp-lsp
uv sync --locked --all-groups
uv run --locked pre-commit install
uv run --locked pre-commit run --all-files
uv build --no-sources
uv run --locked python tools/verify_distribution.pyIf machine uv differs, use uv tool run --from uv==0.13.6 uv ...; ensure nested
hooks resolve uv 0.13.6 too. Do not bypass the project version check.
The four routine hooks run Ruff check/format, ty and unittest, including actual
pinned-ty and outside-checkout installed nine-tool integration tests. Fixture setup
uses the exported runtime lock graph before MCP calls. Packaging checks alone
are not semantic acceptance. Inherited
engineering utility tests run separately, outside the routine gate. Deeper adoption/release
analysis remains outside routine hooks and CI; see the evidence record.
Install the development wheel with
uv pip install --python /path/to/clean/venv/bin/python dist/gz_mcp_lsp-0.1.0-py3-none-any.whl.
The development package declares official mcp==2.3.0 and provides the stdio CLI
gz-mcp-lsp --config /absolute/path/to/settings.json. T6 implements definition, references, diagnostics and hover with exact URI/UTF-16 positions,
complete authorized declaration source and saved-byte generation checks. See
navigation evidence and
pull-diagnostic evidence. See the T4 receipt
and actual Codex checkpoint.
There is no published installation, PyPI release or tag. Local Windows x86_64
and Linux x86_64 gates pass; verify hosted platform results for the exact branch
commit through the Actions link in the handoff.
Related MCP server: lsp-mcp-server
Codex registration (development)
The CLI, JSON schema and all nine tools exist in the development package.
The T11 acceptance record distinguishes
installed/runtime checks, native host approval protocol evidence and platform results.
This example requires a separately installed development wheel.
Provision the adapter and ty before registration; do not use request-time uvx
or another launcher that may fetch dependencies.
[mcp_servers.gz_mcp_lsp]
command = "/absolute/path/to/preinstalled/bin/gz-mcp-lsp"
args = ["--config", "/absolute/path/to/settings.json"]
startup_timeout_sec = 30
tool_timeout_sec = 60
[mcp_servers.gz_mcp_lsp.tools.rename_symbol]
approval_mode = "prompt"
[mcp_servers.gz_mcp_lsp.tools.edit_file]
approval_mode = "prompt"Example application settings.json:
{
"schema_version": 1,
"workspace_root": "/absolute/path/to/workspace",
"ty_executable": "/absolute/path/to/preinstalled/bin/ty",
"request_timeout_seconds": 20
}The settings loader implements this JSON contract without environment-variable
overrides; see configuration. The CLI fixes transport to stdio. Codex's native TOML remains its own contract.
The host supports stdio launch commands in trusted project .codex/config.toml;
see official Codex MCP documentation.
Do not add this development entry to the adopted engineering plugin configuration yet.
For a supported Linux adapter launched by a Windows host, see the
isolated WSL registration example.
Contribution and security
Read CONTRIBUTING.md, SECURITY.md and AGENTS.md. Keep private code, paths, logs and credentials out of public fixtures and issues.
Acknowledgments and license
Isaac Phi's mcp-language-server, reviewed at v0.1.1, informed the idea of exposing language-server navigation through MCP and provides the functional baseline for the revised specification. gz-mcp-lsp is an independent Python implementation; no upstream source has been ported or copied. It is not an official fork or upstream continuation, and is not affiliated with Astral or OpenAI.
This repository was generated directly from
tvproductions/gz-skills-first-starter.
Its adopted published baseline is v0.7.1 at
27594dbaec853e6bde037b0aaf3f0427a3f053f9, recorded in
.skills-first/bundle.lock.json. The inherited policy and analysis tools retain
their provenance and notices. Historical starter documents describe that project,
not delivered adapter behavior; see documentation scope.
Superpowers and gz-skills remain independently owned native plugins; Backplane
is deferred and disabled. The user selected the lite lane.
The project independently selects gz-skills v0.8.3 while retaining the adopted
starter v0.7.1. Its existing skills-first-starter marketplace identity now reads
this repository's project-owned catalog on main; every component in that catalog
has an immutable full commit pin. This reconciles the newer plugin without a
second discovery source or a shared-starter update. See the
dependency-update receipt for
rollback pins, verification and native loading evidence.
Original project work and inherited starter material use the MIT license. Upstream projects and future dependencies retain their own licenses; see NOTICE.
This server cannot be deployed
Maintenance
Related MCP Connectors
Context and control for any AI. Agent identity, scoped access, and proof across tools.
The Remote MCP server acts as a standardized bridge between LLM applications (like Claude, ChatGPT, and Cursor) and external services, enabling AI agents to access external tools and resources. Its primary capability is providing a centralized search tool to discover other MCP servers and their respective tools. Unlike local implementations, it runs remotely with OAuth authentication and permission controls for security.
The Cortex MCP server provides read-only access to real-time engineering context from the Cortex developer portal, allowing AI coding assistants to answer natural language questions about your organization's catalog (microservices, libraries, domains, teams, infrastructure), scorecards (engineering standards and best practices), initiatives (goals and deadlines), and Engineering Intelligence metrics. It includes tools for querying documentation, tracking personal entities, and accessing AI-assisted insights across the entire Cortex ecosystem.
The context platform for AI agents: code search, team decisions, and grounded project context.
Related MCP Servers
- AlicenseNot gradedqualityFmaintenanceBridges the Model Context Protocol with Language Server Protocol to provide AI agents with persistent access to code intelligence features including navigation, diagnostics, refactoring, and completion across 7+ programming languages.1,687 npmMIT
- AlicenseNot gradedqualityDmaintenanceExposes type-aware code navigation and fast file search to AI agents via language servers, enabling definitions, references, symbols, and file lookup without reading entire codebases.1,687 npmMIT
- FlicenseNot gradedqualityAmaintenanceEnables AI coding agents to obtain real-time language-server diagnostics, type information, symbol definitions, references, and workspace-wide renames directly from open editor instances, avoiding slow external type and lint checks.-
- AlicenseNot gradedqualityAmaintenanceEnables MCP-compatible AI coding agents to retrieve real-time language-server diagnostics, type and symbol information, definitions, references, workspace-wide renames, and file operations from an open editor. This provides accurate code navigation and editing context without waiting for slow tsc or eslint checks.MIT