lsp-mcp
# langtools-mcp
> [!WARNING]
> 🚧 This is actively being developed, so expect issues. Currently focusing on compatibility with the on-machine AI agent [Goose](https://block.github.io/goose/docs/quickstart/). 🚧
[](https://opensource.org/licenses/MIT)
**langtools-mcp** is a Model Context Protocol (MCP) server and client toolkit that gives LLMs and AI agents unified access to real static analysis tools—including batch CLI checkers (like Ruff and go vet) and LSPs (like gopls, rust-analyzer, and more).
We've all been there:
- LLM writes code that doesn't compile due to hallucinating standard libraries that don't exist, syntax errors, etc.
- LLM writes functional code, but the style, formatting, and linting is non-existent (I'm looking at you, unused imports)
**langtools-mcp** aims to help solve this by letting your AI and agentic apps **catch, explain, and even fix** issues in code, by calling the same tools expert programmers use. The goal is simply for this MCP to be a tool that the LLM begins using as part of its dev cycle. Just like IDEs and LSPs supercharged humans' ability to quickly assess and fix issues during the dev process, langtools aims to do this with an MCP tool.
- 🧠**Supercharge Agents:** Let your LLMs/AI validate, lint, and debug their own code.
- 🧩 **Modular & Extensible:** Add new languages/tools in minutes via strategies.
- âš¡ **Daemon or Batch:** Runs as a fast HTTP daemon for LSP and batch CLI tools.
---
## Quickstart
### Configuring for Project [Goose](https://block.github.io/goose/docs/quickstart/)
```yaml
langtools:
args:
- --from
- https://github.com/flothjl/langtools-mcp
- langtools-mcp
bundled: null
cmd: uvx
description: null
enabled: true
env_keys: []
envs: {}
name: langtools-mcp
timeout: null
type: stdio
```
<p align="center">
<img src="./assets/goose_sample.png" style="width: 75%;">
</p>
#### Configuration Options
```bash
export LANGTOOLS_PYTHON_TOOLS='["ruff"]'
export LANGTOOLS_GO_GO_TOOLS='["vet"]'
```
| Language | Tools |
| -------- | ------------- |
| Python | ruff, pyright |
| Go | vet |
---
## Installation
```bash
git clone https://github.com/flothjl/langtools-mcp.git
cd langtools-mcp
uv sync # or pip install -e .[dev]
```
**Requirements:** Python 3.10+, plus [ruff](https://docs.astral.sh/ruff/), [pyright](https://github.com/microsoft/pyright), and [Go](https://go.dev/doc/install) for Go support (must be in your PATH).
---
## Roadmap & Supported Tools
- [x] **Python**: Ruff, Pyright (CLI)
- [x] **Go**: go vet (CLI)
- [ ] **Rust**: rust-analyzer (LSP)
- [ ] **JavaScript/TypeScript**: tsc, eslint (planned)
Want to add support for your favorite tool or language?
Open a [PR](https://github.com/flothjl/langtools-mcp/pulls) or start a [Discussion](https://github.com/flothjl/langtools-mcp/discussions)!
---
## Contributing
- Fork, clone, and submit a PR!
- Code and docs welcome for new languages, better error messages, and more.
TDQS
Scored across 1 tool
With only one tool, there is no possibility of ambiguity or overlap between tools. The tool's purpose is clearly defined as analyzing a file, leaving no room for confusion or misselection.
A single tool inherently has perfect naming consistency, as there are no other tools to compare against. The tool name 'AnalyzeFile' follows a clear verb_noun pattern, which would be consistent if more tools were added.
One tool is too few for a server named 'lsp-mcp', which suggests a Language Server Protocol domain. A single analysis tool lacks the expected scope for language features like diagnostics, completions, or refactoring, making it feel incomplete and thin.
The tool set is severely incomplete for an LSP-related server. It only provides file analysis, missing essential operations such as getting diagnostics, providing code completions, or handling references, which are core to language server functionality.