Skip to content
Memexa
FREN
Get started
Persistent memory · MCP protocol

Your agent forgets everything. Memexa remembers.

Decisions, bugs, conventions, context — structured memory that survives every session, wired into Claude Code, Cursor and opencode through one MCP server.

Get started freeDocumentation
  • 20 MCP tools
  • One config block
  • No card to start
~/.cursor/mcp.json
{
  "mcpServers": {
    "memexa": {
      "command": "npx",
      "args": ["-y", "@memexa/mcp-server"],
      "env": { "MEMEXA_API_KEY": "memexa_sk_•••" }
    }
  }
}
# Memory connected.

Works with the agent you already use.

  • Claude Code
  • Cursor
  • opencode
  • Any MCP client
Before / after

The same session, with and without memory.

Tuesday: the agent picks up the auth work started on Monday.

Without Memexa

session-2 · no memory
  • youContinue the auth work from yesterday.
  • agentWhich stack is this? Which auth library did we pick?
  • youFastAPI, JWT in httpOnly cookies. We decided that yesterday.
  • agentWhere did we stop? Re-reading the repository…
  • agentFix: refresh the token on every 401 response.
  • noteThe fix you ruled out yesterday: parallel requests trigger parallel refreshes and revoke the session.

With Memexa

session-2 · memory connected
  • youContinue the auth work from yesterday.
  • callget_context({ task_hint: "auth: refresh tokens" })
  • resultdecision JWT in httpOnly cookies, no localStorage
  • resultbug parallel 401s → parallel refreshes → session revoked
  • resultprogress refresh rotation: in progress
  • agentResuming refresh rotation, with a single refresh in flight.
  • calladd_decision({ decision: "Single-flight refresh", context: "Parallel 401s revoke the session", alternatives: [{ option: "Refresh on every 401", rejected_reason: "Revokes the session" }] })
  • calllog_bug({ description: "Refresh cookie kept after logout", cause: "No Set-Cookie on /logout", solution: "Clear the cookie with Max-Age=0" })

Illustrative session. The entries shown are what an agent reads and writes through the MCP tools.

Features

Structured memory, not a history log.

Four primitives, exposed as MCP tools straight inside your agent.

  • get_context()

    Persistent context

    On session start, the compressed, relevant context is restored automatically.

  • add_decision()

    Tracked decisions

    Every choice stored with its context, alternatives and full reasoning.

  • log_bug()

    Remembered bugs

    Root cause and fix: a bug solved once is never re-debugged.

  • search_cross_project()

    Cross-project search

    Find a pattern or decision shared across projects, semantically.

How it works

Three steps, then never re-contextualize again.

  1. Connect the MCP server

    One config block, your API key. Nothing to host.

  2. The agent writes its memory

    Decisions, bugs and context saved as it works, versioned.

  3. The next session knows it all

    Context restored on startup. The agent picks up where it left off.

Pricing

Two plans. No card to start.

  • Free

    To get started.

    0 XOF

    free, forever

    Get started
    • 2 projects
    • 100 API calls / day
    • 90 days of history
    • All 20 MCP tools
    • Semantic search
  • Recommended

    Pro

    No limits at all.

    5,500 XOF/ mo

    about 9 USD / month

    Go Pro
    • Unlimited projects
    • Unlimited API calls
    • Unlimited history
    • Cross-project search
    • Priority support

Mobile money — Wave, Orange Money, MTN — or card, via GeniusPay.
No card to start. Cancel any time.

FAQ

Frequently asked

A decision, a solved bug, a convention or a progress snapshot. Each entry is versioned and semantically queryable.

Give your agent a memory.

Get started freeDocumentation