Tvix MCP

Explore the MCP tools, resources, and prompts that power Tvix agent integration. Learn how agents store, search, and retrieve Memory and Wiki context through the Model Context Protocol.

Tvix connects to AI agents via the Model Context Protocol (MCP), providing a standard way for agents to store, search, and retrieve Memory and Wiki context.

At a Glance

Type	Name	What it does
Tool	`search_memory`	Hybrid search over personal/work memories
Tool	`search_wiki`	Search wiki notes, resources, and projects
Tool	`capture_memory`	Capture new episodic memory from text
Tool	`create_wiki_note`	Create a wiki note document
Tool	`get_profile`	User profile and core memory blocks
Tool	`get_recent`	Recent memories and inbox captures
Tool	`get_commitments`	Open commitments for the user
Tool	`get_blockers`	Open blockers for the user
Tool	`build_engineer_brief`	Daily engineer brief with priorities
Tool	`draft_follow_up`	Draft a follow-up message (may require approval)
Tool	`assign_memory_project`	Assign a memory to a wiki project
Resource	`execution-twin://profile`	Read-only profile / core blocks
Resource	`execution-twin://recent`	Read-only recent memory timeline
Prompt	`start`	Guidance for memory workflow and safety

You don't need to call these yourself. Your agent handles them based on the conversation.

Memory vs Wiki

Tvix exposes two complementary stores through MCP:

Memory (capture_memory, search_memory): Personal and work context — preferences, decisions, episodes from agents and integrations.
Wiki (create_wiki_note, search_wiki): Knowledge Atlas notes, projects, and resources.

For user questions, agents should generally call both search_memory and search_wiki, then combine the results. See the Knowledge Wiki guide.

MCP Server

Tvix runs a remote MCP server (Streamable HTTP) at:

https://api.tvix.ai/mcp

Locally / Tvix for Mac: http://127.0.0.1:8000/mcp. When MCP_PUBLIC_URL / NEXT_PUBLIC_MCP_PUBLIC_URL is set (OAuth tunnels), the public URL is {that origin}/mcp.

Property	Value
Transport	Streamable HTTP
Authentication	Bearer agent token (most clients); OAuth for ChatGPT / Poke
Path	`/mcp` on the API host

OpenClaw does not use this MCP endpoint — see OpenClaw.

Generate a token and copy client-specific config from the dashboard Agents tab (+ Add Agent). See the Agents guide for per-client setup.

Note: There is no separate mcp.tvix.ai host in this deployment. Production MCP is served from the API domain (api.tvix.ai).

Tools

Memory and search

Tool	Parameters	Notes
`search_memory`	`query` (required), `limit` (default 10)	Hybrid search with citations
`capture_memory`	`content` (required), `source_type` (default `agent`)	Write tool; may require approval in dashboard mode
`get_recent`	`limit` (default 15)	Recent memories + inbox
`assign_memory_project`	`memory_id` (required), `project_id` (optional)	Link a memory to a wiki project

Wiki

Tool	Parameters	Notes
`search_wiki`	`query` (required), `limit` (default 20)	Notes, resources, projects
`create_wiki_note`	`name` (required), `content` (optional)	Creates a Knowledge Atlas wiki note

There are currently no MCP tools named add_memory, add_wiki, update_wiki, delete_wiki, or get_current_date. Edit or delete wiki notes from the dashboard Wiki UI (or workspace APIs).

Attention / briefs

Tool	Parameters	Notes
`get_profile`	(none)	Profile + core memory blocks
`get_commitments`	(none)	Open commitments
`get_blockers`	(none)	Open blockers
`build_engineer_brief`	(none)	Engineer daily brief
`draft_follow_up`	`context` (required), `channel` (default `slack`)	Draft for human approval

Resources

Resource	URI	Description
User Profile	`execution-twin://profile`	Profile and core blocks (same data as `get_profile`)
Recent	`execution-twin://recent`	Recent memories / inbox (same data as `get_recent`)

The execution-twin:// URI scheme is the current resource identifier in the hosted MCP server (legacy internal name). Prefer the matching tools if a client does not surface resources.

Prompts

Prompt	Description
start	Teaches the agent when to search Memory vs Wiki, when to capture, and what to avoid storing

Recommended Agent Behavior

For questions about saved knowledge, search Memory and Wiki together when both could matter.
Save durable personal context with capture_memory; skip secrets and one-off transient data unless explicitly requested.
Save factual / reference notes with create_wiki_note.
Use get_profile / execution-twin://profile for stable user settings.
Use get_recent / execution-twin://recent for latest timeline context.

Connection Setup

For detailed setup instructions for each agent, see the Agents guide.

Next Steps

Memory

Explore and manage memories.

Knowledge Wiki

Knowledge Atlas notes, projects, and inbox.

Chat in Dashboard

Talk to your knowledge base without an external agent.