SPM — Claude Code MCP server for project memory
A hosted MCP server that keeps requirements, decisions and tests in one place and hands them to your agent in the next chat.
claude mcp add --transport http spm-structured-project-memory https://getspm.com/v1/mcpWhat SPM - Structured Project Memory does
SPM stores project knowledge as memory that survives across chats, runs and tools. You point a client at the remote endpoint https://getspm.com/v1/mcp, and the agent can resolve which project is active, pull the context it needs, then report decisions, tests and evidence back when the work is done.
Memory comes in scoped packs with provenance and expiry. An agent can verify hashes, query temporal state, run policy-aware preflight checks and inspect the capture journal. Pulling context from several projects at once is not automatic — the user has to ask for it, because each task keeps a single active project. Moving to another agent or device goes through a one-time handoff that carries references instead of memory bodies or credentials.
Before anything works you need an SPM account, an organization with at least one project, a trial or paid plan, and a token scoped to one project, a chosen set, or everything the signed-in user may access. Keep that token in an environment variable or the client's secret storage. Committing it to a repo is the wrong move, since tokens are meant to be revocable from SPM. The repository ships native bundles for Codex, Claude, Cursor and OpenClaw, plus authorization helpers and config examples.
What it helps with
- carry project memory into a fresh chat without retyping it
- hand a contractor a bounded slice of a project with sources attached
- check where an agent got its context and whether the hashes match
- record decisions and test results once a task finishes
- resume the same project in another agent or on another machine
Example prompts for the agent
- Show the current requirements and decisions for the active project
- Build a context pack for the
billingproject and verify its provenance - Report the finished task and passing tests back to SPM
How to install SPM - Structured Project Memory
Platform CLI: remote server
claude mcp add --transport http spm-structured-project-memory https://getspm.com/v1/mcpPlatform CLI: remote server
codex mcp add spm-structured-project-memory --url https://getspm.com/v1/mcpConfig file
{
"mcpServers": {
"spm-structured-project-memory": {
"url": "https://getspm.com/v1/mcp"
}
}
}Config file
{
"servers": {
"spm-structured-project-memory": {
"type": "http",
"url": "https://getspm.com/v1/mcp"
}
}
}Platform CLI: remote server
gemini mcp add -t http spm-structured-project-memory https://getspm.com/v1/mcpStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about SPM - Structured Project Memory
How do I install SPM in Claude Code?
Run claude mcp add --transport http spm-structured-project-memory https://getspm.com/v1/mcp. You still need to authorize a token for the project you want to use.
What does SPM do?
It keeps project memory alive between agent sessions: requirements, current decisions, completed work and source-backed context. The agent requests the slice it needs and reports decisions and evidence back afterwards.
Can an agent mix context from several projects?
Only when the user explicitly asks. A task keeps one active project by default, even if the token also covers others in the organization.
