Titentiten.devInstall
The connection guide

An MCP memory server for durable agent context.

An MCP memory server gives compatible agents a standard tool interface for durable memory. Titen serves seven tools over one authenticated HTTP endpoint and keeps the full REST API available when a workflow needs more control.

POST /mcp
one authenticated HTTP endpoint
7 tools
memory, context, state, coordination
3 revisions
supported MCP protocol versions
01

What does an MCP memory server do?

MCP lets an agent host discover and call tools through a standard JSON-RPC handshake. The host does not need a Titen-specific client library to remember an observation, compile context, save a checkpoint, or hand work to another principal.

The endpoint still needs a security boundary. Titen uses the same bearer keys as REST and requires the mcp:call scope. Each agent should receive a separate key with only the authority it needs.

Read the complete MCP contract →
02

Install and connect Titen

Install the CLI, create the local store, and start the Bun/SQLite runtime. The MCP endpoint then listens beside the REST API.

curl -fsSL https://titen.dev/install.sh | bash
titen bootstrap --org 'My Org'
titen serve

# MCP endpoint
http://127.0.0.1:8787/mcp
  • Send the bearer key in the Authorization header.
  • Keep keys in the host secret or environment configuration, never in a repository file.
  • Grant mcp:call deliberately because it includes write-capable tools.
Choose an installer →
03

The seven MCP memory tools

The tool list is intentionally small. It covers the memory loop and the minimum coordination state an agent team needs.

  • titen_remember appends an observation.
  • titen_compile builds a bounded context pack.
  • titen_feedback records whether recalled context helped.
  • titen_checkpoint_save and titen_checkpoint_get preserve expiring task state.
  • titen_lease_acquire prevents silent duplicate work when agents cooperate.
  • titen_handoff transfers a subject to another principal.
Understand checkpoint state →
04

When to use MCP and when to use REST

Use MCP for a portable tool connection and short memory operations. Use REST when the workflow needs semantic retrieval, full provenance metadata, project-scoped compilation, idempotency, events, or stricter trust enforcement.

The current MCP compile path is lexical-only, even when sqlite-vec is available. MCP remember also writes fewer lifecycle rows than the REST observation route. These differences are part of the contract, not implementation trivia.

A practical setup can use both: MCP for agent-host interoperability and REST for load-bearing writes or context compilation that needs the complete policy surface.

Compare the REST API →