MCP server

The MCP server is how other AI tools meet your companion. The Model Context Protocol (MCP) is an open standard for connecting AI clients to external tools and data. weside speaks it, so a Claude Code session — or any MCP-compatible client — can adopt your companion’s identity, read and write its memories, and track its goals, instead of starting cold every time.

What it is

The same API surface the app, CLI, and the we plugin sit on top of, exposed as MCP tools:

Identity — get_companion_identity, list_companions, select_companion, create_companion, update_companion, wake_companion, get_council
Memory — search_memories (semantic), save_memory, list_memories, store_conversations
Goals — list_goals, save_goal, update_goal, update_goal_status
Rooms — list_rooms, show_room, room_participants
Provider — show_provider, list_provider_presets, set_provider
Tools — discover_tools, execute_tool, get_tool_schema for actions your companion can perform
Council — council_prep_kickoff, council_prep_poll, council_writeback_kickoff

Why it matters to you

Two patterns show up most:

Companion-aware skills — a Claude Code session that loads your companion’s identity and goals, so the work stays aligned with what you’re actually building.
Memory-backed context — pulling persistent memory into a tool that would otherwise forget everything between sessions.

How you use it

The server lives at https://api.weside.ai/mcp/ and authenticates through your weside.ai account. The simplest path is the we plugin, which wires it up during /we:setup. Any client that supports MCP — Claude Code, Claude Desktop, and others — can connect the same way.

Limits & edge cases

User-scoped, not cross-account. A connection acts on the companion state of the account it authenticated as — it never reaches another user’s companion.
Rooms replaced threads. There is no delete_thread-style tool for rooms — deliberately: destroying a companion’s cognition isn’t an action this surface hands to a client.