Home/Documentation

⌘K

MCP Server

Model Context Protocol5 min setup

Connect MemoAir to Cursor, Claude, or any MCP-compatible AI tool. Your assistant gets persistent memory that survives across every conversation.

Connect Cursor — 5 minutes

Cursor natively supports HTTP MCP transport. The fastest path uses the dashboard one-click install:

One-click install (recommended)

1.Open dashboard.memoair.space
2.Go to Connected Apps → click Cursor → Install
3.Cursor opens and configures MCP automatically
4.Run npx -p memoair-mcp memoair-init in your project root
5.Press Cmd+Shift+P → Developer: Reload Window

Terminal fallback — if the deeplink didn't open, this single command writes .cursor/mcp.json and runs memoair-init together:

BASH

npx -p memoair-mcp memoair-connect cursor \
  --mcp-url "https://mcp-server.memoair.space/mcp" \
  --project-api-key "memoair_pk_your_project_key" \
  --user-id "your-email@example.com" \
  --init

Manual config — create .cursor/mcp.json in your project root:

.cursor/mcp.json

JSON

{
  "mcpServers": {
    "memoair": {
      "url": "https://mcp-server.memoair.space/mcp",
      "headers": {
        "X-API-Key": "memoair_pk_your_project_key",
        "X-User-ID": "your-email@example.com"
      }
    }
  }
}

Smart rules: memoair-init installs a Cursor rule that auto-retrieves only the coding rules relevant to your current task — saving 60–80% of rule tokens per request. Full Cursor guide →

Verify your setup

Press Cmd+Shift+P → MCP: List Servers — confirm memoair appears. Then ask Cursor: "Use MemoAir to save that this project uses MemoAir for memory." — if MemoAir responds, you're done.

Other Clients

Prerequisites

Node.js 20+ (required by the memoair-mcp package)
A MemoAir project API key — prefix memoair_pk_ — from your dashboard settings
Cursor, Claude Desktop, or Claude Code

Available Tools

Once connected, your AI assistant has access to these memory tools:

search_memories

Search for coding conventions, architecture context, and project knowledge. Pass file_path to auto-scope results to the relevant service.

query (required)limitviewtagstag_modefile_pathinclude_supersededsession_id

save_memory

Save a coding insight, convention, or pattern to working memory. Include relevant tags to make it findable in future sessions.

content (required)typesource_urlsource_titleimage_urldescriptiontags

save_document

Save or update a versioned canonical document — architecture decisions, coding rules, conventions, or setup guides. Old versions are automatically superseded.

content (required)doc_family_id (required)source_titlesource_urltagssupersedes_doc_family_id

list_tags

Discover what knowledge categories exist in the project. Returns tags grouped by namespace (service, type, topic, lang).

prefix

delete_memory

Delete a specific memory by its ID, obtained from search results.

id (required)

update_memory

Update an existing memory with new content.

id (required)content (required)typesource_urlsource_titleimage_urldescription

analyze_query

Save new information and search for related context simultaneously. Use when the user shares new facts, updates, or corrections.

query (required)source_urlsource_title

list_working_memory

List items in working memory grouped by category (tab, clip, task, agent, chat). Useful for cross-agent context sharing.

categoriesinclude_contentlimit

get_working_memory_context

Retrieve context from a specific working memory item or search within a category.

category (required)queryitem_idlimit

get_doc_family

Retrieve the current version of a canonical document by family ID. More precise than search for known doc families.

doc_family_id (required)include_history

get_recent_deltas

Get recent memory deltas for a doc family or service. Use after get_doc_family to catch up on decisions since the last canonical save.

doc_family_idservice_tagsince_doc_familylimit

Server Prompts

MCP clients that support prompts can invoke these built-in workflows:

wiki_coding_assistant

Initialize as a wiki-first coding assistant: load project guides before searching, save deltas after changes.

load_project_context

Load project context using the guide-first pattern: get_doc_family → get_recent_deltas → search_memories for gaps.

record_coding_decision

Record a coding decision or learning — routes to save_document for full docs, save_memory for small deltas.

Environment Variables

For stdio mode or self-hosted deployments:

Variable	Required	Description
MEMOAIR_API_KEY	stdio only	Project API key (prefix `memoair_pk_`). HTTP mode uses the `X-API-Key` header instead.
MEMOAIR_USER_ID	Optional	User ID for memory isolation in stdio mode. HTTP mode uses the `X-User-ID` header.
MEMOAIR_API_BASE	Optional	Backend URL (defaults to `http://localhost:8080`). Set for self-hosted instances.
MEMOAIR_TRANSPORT	Optional	Transport mode: `http` (default for hosted), `stdio` (local dev).

Troubleshooting

Cursor deeplink did not open

Use the terminal fallback: npx -p memoair-mcp memoair-connect cursor --mcp-url ... --project-api-key ... --user-id ... --init

MCP server not appearing in tool list

Run Developer: Reload Window from the Command Palette. If still missing, verify .cursor/mcp.json exists and contains a valid mcpServers object.

Authentication errors

Verify your API key starts with memoair_pk_ and hasn't expired. Regenerate from dashboard settings if needed.

Node.js version error

memoair-mcp requires Node.js 20+. Run node --version to check, then upgrade via nvm or nodejs.org if needed.

.cursor/mcp.json has invalid JSON

Open the file and validate it. The file must be a JSON object with a top-level mcpServers key. Remove comments or trailing commas.

memoair-init did not create files

Existing files are intentionally preserved. Check that .cursor/rules/memoair.mdc and memoair.md don't already exist. Delete them if you want a fresh install.

Full Cursor Integration Guide

Smart rules, token savings, project guide, and team workflows