import { z } from "zod";
import type { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { executeMemvid, buildArgs } from "../executor.js";
import { filePathSchema, frameIdSchema, inputPathSchema, formatToolResult, TIMEOUTS, ANNOTATIONS } from "../types.js";
export function registerWriteTools(server: McpServer) {
server.registerTool(
"memvid_put",
{
title: "Add Content",
description: `Add content to a memory file from a file or directory.
Supports multiple file types: text, markdown, code, PDF, images (with OCR).
Use embed=true to generate vector embeddings for semantic search.
Args:
file: Path to the .mv2 memory file
input: Path to file or directory to ingest
recursive: Include subdirectories when input is a directory
parallel: Process files in parallel for faster ingestion
embed: Generate vector embeddings (requires embedder.toml configuration)
log: Path to log file for detailed operation logging
Returns:
{
"files_added": number,
"frames_created": number,
"embeddings_generated": number (if embed=true)
}
Examples:
- Add single file: input="/docs/readme.md"
- Add directory: input="/docs", recursive=true
- With embeddings: input="/docs", embed=true`,
inputSchema: z.object({
file: filePathSchema,
input: inputPathSchema,
recursive: z.boolean().optional().describe("Include subdirectories recursively"),
parallel: z.boolean().optional().describe("Enable parallel processing"),
embed: z.boolean().optional().describe("Generate embeddings for vector search (requires embedder.toml config)"),
log: z.string().optional().describe("Log file path for detailed operation logging")
}).strict(),
annotations: ANNOTATIONS.WRITE
},
async ({ file, input, recursive, parallel, embed, log }) => {
const args = buildArgs(["put", "--input", input, file], { recursive, parallel, embed, log });
const result = await executeMemvid(args, { timeout: TIMEOUTS.HEAVY });
return formatToolResult(result);
}
);
server.registerTool(
"memvid_put_many",
{
title: "Batch Add Content",
description: `Batch add multiple files from a directory with progress tracking.
Optimized for large directories with commit batching.
Args:
file: Path to the .mv2 memory file
input: Path to input directory
recursive: Include subdirectories
parallel: Process files in parallel
batch_size: Number of files per commit batch (default: 100)
Returns:
JSON with batch processing statistics`,
inputSchema: z.object({
file: filePathSchema,
input: inputPathSchema.describe("Input directory path"),
recursive: z.boolean().optional().describe("Include subdirectories recursively"),
parallel: z.boolean().optional().describe("Enable parallel processing"),
batch_size: z.number().int().positive().optional().describe("Batch size for commits")
}).strict(),
annotations: ANNOTATIONS.WRITE
},
async ({ file, input, recursive, parallel, batch_size }) => {
const args = buildArgs(["put-many", "--input", input, file], { recursive, parallel, batch_size });
const result = await executeMemvid(args, { timeout: TIMEOUTS.HEAVY });
return formatToolResult(result);
}
);
server.registerTool(
"memvid_view",
{
title: "View Frame",
description: `View content of a specific frame by ID.
Args:
file: Path to the .mv2 memory file
frame_id: Frame ID (non-negative integer)
raw: Show raw content without formatting/highlighting
Returns:
{
"frame_id": number,
"content": string,
"uri": string,
"created_at": string,
"metadata": object
}`,
inputSchema: z.object({
file: filePathSchema,
frame_id: frameIdSchema,
raw: z.boolean().optional().describe("Show raw content without formatting")
}).strict(),
annotations: ANNOTATIONS.READ_ONLY
},
async ({ file, frame_id, raw }) => {
const args = buildArgs(["view", file, String(frame_id)], { raw });
const result = await executeMemvid(args);
return formatToolResult(result);
}
);
server.registerTool(
"memvid_update",
{
title: "Update Frame",
description: `Update content of a specific frame.
Replaces the entire content of the frame. For partial corrections, use memvid_correct instead.
Args:
file: Path to the .mv2 memory file
frame_id: Frame ID to update
content: New content for the frame
Returns:
JSON with update confirmation`,
inputSchema: z.object({
file: filePathSchema,
frame_id: frameIdSchema,
content: z.string().min(1).describe("New content for the frame")
}).strict(),
annotations: ANNOTATIONS.DESTRUCTIVE
},
async ({ file, frame_id, content }) => {
const args = ["update", file, String(frame_id), "--content", content];
const result = await executeMemvid(args);
return formatToolResult(result);
}
);
server.registerTool(
"memvid_delete",
{
title: "Delete Frame",
description: `Delete a specific frame from memory.
This operation is destructive and cannot be undone.
Args:
file: Path to the .mv2 memory file
frame_id: Frame ID to delete
force: Skip confirmation prompt
Returns:
JSON with deletion confirmation`,
inputSchema: z.object({
file: filePathSchema,
frame_id: frameIdSchema,
force: z.boolean().optional().describe("Force deletion without confirmation")
}).strict(),
annotations: { ...ANNOTATIONS.DESTRUCTIVE, idempotentHint: true }
},
async ({ file, frame_id, force }) => {
const args = buildArgs(["delete", file, String(frame_id)], { force });
const result = await executeMemvid(args);
return formatToolResult(result);
}
);
server.registerTool(
"memvid_correct",
{
title: "Correct Frame",
description: `Correct/amend content of a frame.
Creates a correction record preserving the original content for audit purposes.
Use this instead of update when you want to maintain history.
Args:
file: Path to the .mv2 memory file
frame_id: Frame ID to correct
content: Corrected content
Returns:
JSON with correction record ID`,
inputSchema: z.object({
file: filePathSchema,
frame_id: frameIdSchema,
content: z.string().min(1).describe("Corrected content")
}).strict(),
annotations: ANNOTATIONS.WRITE
},
async ({ file, frame_id, content }) => {
const args = ["correct", file, String(frame_id), "--content", content];
const result = await executeMemvid(args);
return formatToolResult(result);
}
);
server.registerTool(
"memvid_api_fetch",
{
title: "Fetch from URL",
description: `Fetch content from a URL and add to memory.
Supports web pages, APIs, and document URLs.
Args:
file: Path to the .mv2 memory file
url: URL to fetch content from
title: Custom title for the fetched content (optional)
Returns:
JSON with fetch status and created frame ID`,
inputSchema: z.object({
file: filePathSchema,
url: z.string().url().describe("URL to fetch content from"),
title: z.string().optional().describe("Custom title for the fetched content")
}).strict(),
annotations: ANNOTATIONS.NETWORK
},
async ({ file, url, title }) => {
const args = buildArgs(["api-fetch", file, url], { title });
const result = await executeMemvid(args);
return formatToolResult(result);
}
);
}
