MCP Tools

MCP Tools (11)

The same tools are available to the chat UI, Cursor, Windsurf, or any MCP-compatible agent.

Tool	Description
remember	Save a memory + auto-generate embedding
recall	Semantic search with similarity scores, full-text fallback
update_memory	Edit by UUID — regenerates embedding if content changes
forget	Delete by UUID
backfill_embeddings	Generate embeddings for memories that don't have one
dedup_memories	Find near-duplicates via cosine similarity — merge or report
read_file	Read any file from disk (max 500 lines, safe extensions only)
scan_project	Scan a folder tree, read key files, infer project context
fetch_url	Fetch a URL, strip HTML, truncate at 15k chars
get_stats	Memory count, type distribution, embedding coverage
search_by_tag	Filter memories by one or more tags

Available file tools

Tool	What it does
read_file	Read any file from disk (max 500 lines by default)
write_file	Overwrite a file completely
append_file	Add content to the end of a file
scan_project	Scan a folder tree up to 3 levels deep

Raising the line limit

The default cap is 500 lines. To increase it, find this in mcp/index.js:

const READ_FILE_CHUNK_SIZE = 500;   // max lines per read_file call
const READ_FILE_MAX_OFFSET = 10_000; // safety ceiling for chunked reads

NOTE: If you ask it to read a large file it'll truncate. For big files you'd either need to raise that limit or use scan_project first to find the right file, then read_file on the specific section you need.

Change it to whatever your use case needs.

Example prompts that just work

• "Read my server.js and tell me what the WebSocket handler does"
• "Scan my project and give me an overview of the structure"
• "Read my .env.example and tell me which variables I still need to fill in"
• "Append a TODO comment to the bottom of mcp/index.js"

[Back to top ↑]

---