Exposes TypeScript LSP-like functionality including type inspection, definition lookup, references finding, hover documentation, diagnostics, and type testing capabilities for TypeScript projects.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@TypeScript LSP MCP get the type at line 15 column 8 in src/user.ts"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
MCP server exposing TypeScript LSP-like functionality to AI agents.
Gives AI agents the same "what's the type at this position?" powers that IDE users have.
# Install to current project (recommended)
npx ts-lsp-mcp install cc
# Or install globally for all projects
npx ts-lsp-mcp install cc --globalThat's it! The MCP server is now available to Claude Code.
npx ts-lsp-mcp uninstall ccRelated MCP server: VSCode MCP
npm install -g ts-lsp-mcp
ts-lsp-mcp install ccAdd to your Claude Code MCP config (.mcp.json in project or ~/.claude/settings.json globally):
{
"mcpServers": {
"ts-lsp-mcp": {
"type": "stdio",
"command": "npx",
"args": ["-y", "ts-lsp-mcp", "serve", "--stdio"]
}
}
}For remote clients or debugging:
ts-lsp-mcp --http --port 3000Endpoints:
GET /sse - SSE connection
POST /message - Send messages
GET /health - Health check
Tool | Description |
| Get the TypeScript type at a specific file:line:col |
| Go to definition |
| Find all references |
| Get hover documentation |
| Get autocomplete suggestions |
| Get type errors and warnings |
| Trace where a type comes from and how it's composed |
| Run type assertions from |
| Type-check inline TypeScript without creating files |
Add type assertions to your code:
// @ts-lsp-mcp expect-type: string
const name = user.name;
// @ts-lsp-mcp expect-type: User
const user = createUser({ name: 'Alice' });
// @ts-lsp-mcp expect-error: 2322
const bad: number = "oops"; // Should have error 2322Run tests:
runTypeTests({ file: "src/types.ts" })
runTypeTests({ pattern: "**/*.type-test.ts" })Supports unified file:line:col format:
{
"tool": "getTypeAtPosition",
"arguments": {
"file": "src/user.ts:10:5"
}
}Or separate parameters:
{
"tool": "getTypeAtPosition",
"arguments": {
"file": "src/user.ts",
"line": 10,
"col": 5
}
}Response:
{
"type": "User",
"expanded": "{ id: number; name: string; email: string }",
"symbol": "user",
"kind": "variable"
}{
"tool": "getDiagnostics",
"arguments": {
"file": "src/user.ts"
}
}{
"tool": "checkInlineCode",
"arguments": {
"code": "const x: number = 'bad';"
}
}Response:
{
"valid": false,
"diagnostics": [{
"line": 1,
"col": 7,
"code": 2322,
"message": "Type 'string' is not assignable to type 'number'."
}]
}Multi-project support: Works with monorepos with multiple tsconfigs
Auto-discovery: Finds tsconfig.json automatically
Smart file resolution: Accepts absolute, relative, or unique filenames
Virtual files: Type-check unsaved code with the content parameter
Efficient: Long-lived daemon caches TypeScript projects
ts-lsp-mcp [command] [options]
Commands:
serve Start the MCP server (default)
install <target> Install into an AI assistant (cc, claude-code, claude)
uninstall <target> Uninstall from an AI assistant
Serve options:
--stdio Use stdio transport (default)
--http Use HTTP/SSE transport
--port <port> HTTP server port (default: 3000)
--host <host> HTTP server host (default: 127.0.0.1)
--debug Enable debug logging
Install options:
--global Install globally (user scope) instead of project
--name <name> Custom name for the MCP server (default: ts-lsp-mcp)
General:
-V, --version Output version number
-h, --help Display helpMIT
Type-check inline TypeScript code without creating a file
Inputs: code (required), fileName, tsconfig.
Get autocomplete suggestions at a position
Inputs: file (required), col, line, prefix, +3 more.
Get the definition location(s) for a symbol at a position (go-to-definition)
Inputs: file (required), col, line, content, +1 more.
Get TypeScript diagnostics (errors, warnings) for a file or project
Inputs: file, content, severity, projectRoot.
Get hover information (documentation, type signature) at a position
Inputs: file (required), col, line, content, +1 more.
Find all references to a symbol at a position
Inputs: file (required), col, line, content, +3 more.
Get the TypeScript type at a specific position in a file
Inputs: file (required), col, line, content, +2 more.
Run type test assertions from files with @ts-lsp-mcp expect-type/expect-error comments
Inputs: file, pattern, projectRoot.
Trace where a type comes from and how it is composed
Inputs: file (required), col, line, depth, +2 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 9 tools
Each tool targets a distinct LSP feature (type-checking, completions, definitions, diagnostics, hover, references, type queries, type tests, type tracing). There is no overlap or ambiguity between tools.
Most tools follow a 'get' + noun pattern (getCompletions, getDefinition, etc.), with two exceptions: 'checkInlineCode' uses 'check' and 'runTypeTests' uses 'run'. This minor inconsistency prevents a perfect score.
9 tools is an appropriate size for a TypeScript LSP server. It covers essential features without being bloated or too sparse.
The tools cover core LSP operations (diagnostics, hover, definition, references, completions, type queries) and even include niche features (type tracing, type tests). Missing common operations like code actions, rename, or formatting, but the set is largely complete for its apparent purpose.
Coding agents in multi-service codebases routinely rebuild existing helpers, trust stale type definitions, and modify API contracts without knowing who consumes them. Carrick solves this by indexing your entire TypeScript ecosystem across service and repository boundaries. By integrating deeply with the TypeScript compiler, Carrick traces every route, type, and cross-service call while recording function behaviour so agents search by intent rather than name. Delivered via MCP for AI agents and LSP for IDEs, Carrick ensures models see existing endpoints and utilities before generating new code. The scanner is source-available and runs from your CLI or CI pipeline.
Web scraping, Google SERP, Markdown and ChatGPT/Gemini answers as typed tools for AI agents.
Write, compile and share Typst documents (papers, slides, drawings) from any AI agent.
Search @imqueue docs and scaffold typed services & clients from your AI coding agent.
