@dreki-gg/pi-lsp

Language-agnostic code intelligence for pi via Language Server Protocol. Purely config-driven — you define which servers to use per project.

Install

pi install npm:@dreki-gg/pi-lsp

Tool

Single unified lsp tool with 11 operations:

| Operation | Description | Required params | |-----------|-------------|-----------------| | diagnostics | Type errors + lint warnings | filePath | | hover | Type info and documentation | filePath, line, character | | goToDefinition | Find where a symbol is defined | filePath, line, character | | findReferences | Find all references to a symbol | filePath, line, character | | goToImplementation | Find implementations of interface/abstract | filePath, line, character | | documentSymbol | List all symbols in a file | filePath | | workspaceSymbol | Search symbols across the workspace | query | | prepareCallHierarchy | Get call hierarchy item at position | filePath, line, character | | incomingCalls | Find callers of a function | filePath, line, character | | outgoingCalls | Find callees of a function | filePath, line, character | | codeActions | Quick fixes and refactoring suggestions | filePath, line, character |

All line/character params are 1-indexed (matching the read tool output).

Configuration

Servers are configured via two config files (project overrides global):

| File | Scope | |------|-------| | ~/.pi/agent/extensions/lsp/config.json | Global defaults | | .pi/lsp.json | Project-local overrides |

Example: TypeScript + oxlint

.pi/lsp.json:

{
  "lsp": {
    "typescript": {
      "command": ["typescript-language-server", "--stdio"],
      "extensions": [".ts", ".tsx", ".js", ".jsx", ".mjs", ".cjs", ".mts", ".cts"]
    }
  }
}

typescript-language-server automatically uses the project's local node_modules/typescript, so your project's TS version is always respected.

Adding other servers

.pi/lsp.json:

{
  "lsp": {
    "rust": {
      "command": ["rust-analyzer"],
      "extensions": [".rs"]
    },
    "python": {
      "command": ["pyright-langserver", "--stdio"],
      "extensions": [".py"],
      "initialization": {
        "python": { "analysis": { "typeCheckingMode": "basic" } }
      }
    }
  }
}

Disabling a server

{
  "lsp": {
    "typescript": { "disabled": true }
  }
}

Disabling all LSP

{
  "lsp": false
}

Server config options

| Property | Type | Description | |----------|------|-------------| | command | string[] | Command + args to spawn (e.g. ["rust-analyzer"]) | | extensions | string[] | File extensions with leading dot | | disabled | boolean | Disable this server | | env | object | Environment variables for the server process | | initialization | object | Options sent during LSP initialize handshake |

How it works

• Auto-detection: Servers are matched to files by extension. Multiple servers can handle the same extension.
• Routing: diagnostics aggregates from all matching servers. Other operations use the first server with the required capability.
• Lazy start: Servers spawn on first tool use, stay alive for the session.
• Config merge: Project .pi/lsp.json overrides global ~/.pi/agent/extensions/lsp/config.json.

Commands

| Command | Description | |---------|-------------| | /lsp | Show server status, detected servers, and extensions | | /lsp-restart | Stop all servers (reinitialize on next tool use) |

Architecture

extensions/lsp/
├── index.ts       — Entry point, server manager, lifecycle
├── protocol.ts    — JSON-RPC over stdio transport
├── client.ts      — High-level LSP client (all 11 operations)
├── config.ts      — Config loading, merging, server resolution
├── tools.ts       — Single unified `lsp` tool registration
├── formatting.ts  — Format all LSP responses for LLM consumption
└── types.ts       — LSP protocol types, config types, operation enums