ExplicitMem MCP

Memory that survives across agent sessions

ExplicitMem MCP stores named memory objects and retrieves them in later agent sessions with their source records attached.

Included with an active usage plan

Available through the hosted catalog. Calls count against the shared allowance on the account plan.

Compare usage plans

Why it exists

Store named memory objects and retrieve them in later agent sessions without relying on chat history that disappears between sessions.

Generic terminals, provider APIs, and custom prompts can cover parts of the work, but the caller must supply permission limits, output requirements, failure handling, and usage accounting.

Connection and contract

Connect from an MCP-aware client with an account-scoped Monarchic key. The hosted route checks access, runs the named product contract, and records the operation on the account receipt.

The product-specific schema defines the inputs and outputs used by clients and downstream automation.

Current limits

The caller supplies workspace and scope IDs. They are not yet derived from the authenticated tenant, so use non-sensitive evaluation memory only.

Needs an authenticated, entitled account and a write-enabled ExplicitMem tool profile. No external model credential is required for the default local-hash path.

What you get

Persist facts, preferences, goals, and working context
Retrieve bounded context with source records
Audit memory objects and changes

Core use case

Store named memory objects and retrieve them in later agent sessions without relying on chat history that disappears between sessions.

Outcomes

Persist facts, preferences, goals, and project context
Retrieve relevant context with the source memory objects
Audit what an agent remembered and how it changed

Primary workflows

Write memory Search memory Retrieve context Audit objects

Research

Published benchmark

How accurately do fixed ExplicitMem retrieval and synthesis configurations support answers across the published evaluation sets?

Use dated, versioned artifacts for LongMemEval-S, LoCoMo, cross-dataset retrieval policy, and generic-answer support checks.

Each result applies only to its named dataset, configuration, metric, and artifact. The studies do not establish a cross-product ranking, general agent accuracy, or hosted latency guarantee.

Read benchmark

Concrete workflow / priced call

Remember a release rule, then retrieve it with provenance

The request, exact tool-call payload, returned output, and usage receipt are shown below.

Request

“Remember that production releases require two reviewers, then prepare the approval constraints.”

Execution

Tool call payload

[
  {
    "tool": "memory.write_constraint",
    "arguments": {
      "workspaceId": "demo-workspace",
      "scopeId": "project_specific",
      "content": "Production releases require two reviewers."
    }
  },
  {
    "tool": "memory.retrieve_context",
    "arguments": {
      "workspaceId": "demo-workspace",
      "scopeId": "project_specific",
      "task": "Prepare the release approval constraints",
      "mode": "task_brief",
      "limit": 5,
      "tokenBudget": 300
    }
  }
]

Writes one explicit constraint, retrieves a bounded context packet, and attaches the source event and deterministic context receipt.

Output

Representative contract excerpt

{
  "write": {
    "eventId": "event-082",
    "objectId": "constraint-014",
    "action": "created",
    "objectType": "constraint",
    "canonicalText": "Production releases require two reviewers"
  },
  "retrieve": {
    "task": "Prepare the release approval constraints",
    "mode": "task_brief",
    "retrievalMode": "context_pack",
    "objectIds": [
      "constraint-014"
    ],
    "contextMarkdown": "Task Context Brief\nTask: Prepare the release approval constraints\n\nAlways-Include Facts:\n- None\n\nFacts:\n- None\n\nHypotheses:\n- None\n\nConstraints:\n- [constraint] Production releases require two reviewers (confidence=0.65)\n\nContested Memory:\n- None\n\nContradictions:\n- None\n\nResolutions:\n- None",
    "directEvidence": [
      {
        "objectId": "constraint-014",
        "objectType": "constraint",
        "text": "Production releases require two reviewers",
        "sourceEventIds": [
          "event-082"
        ],
        "rank": 1,
        "score": 0.4503511301977579,
        "confidence": 0.65,
        "epistemicStatus": "observed",
        "reasons": [
          "semantic=0.24",
          "lexical=0.20",
          "type_boost=+0.07",
          "recency=+0.05"
        ]
      }
    ],
    "answerNotes": [
      {
        "text": "Constraint evidence: Production releases require two reviewers",
        "evidenceObjectIds": [
          "constraint-014"
        ],
        "sourceEventIds": [
          "event-082"
        ]
      }
    ],
    "contextReceiptHash": "sha256:…"
  }
}

Usage rating

At runtime

Recorded on the operation receipt

Each call receives its own measured receipt. Allowance quantities remain unpublished until the fully allocated rate card is frozen.

memory.write_constraint
memory.retrieve_context