FIG. 1 — MCP MEMORY SERVER RENDERED

Persistent, inspectable memory for your agents — in any agent, via MCP.

Connect Answer Engine as an MCP server so every agent answers from your own tenant-isolated sources, with citations, lineage, and an inspectable memory trail.

Answer Engine gives AI agents persistent memory that is grounded in your own sources, exposed through MCP, and inspectable when an answer needs proof. Install the MCP server, connect your library, and Claude Code, Cursor, Codex, Gemini CLI, or any MCP client can recall the same tenant-isolated context.

claude mcp config package @answer-engine/mcp-server

{ "mcpServers": { "answer-engine": { "command": "npx", "args": ["answer-engine-mcp"] } } }
npx answer-engine-mcp
Available now Tenant-isolated grounded Q&A over your sources16 MCP tools over stdio and HTTPInspectable memory lineageNo training on your data
Shipping next Reranked retrieval Self-driving evals + reproducible scorecard Org-shared mode with per-doc ACL Continuous-sync scheduler
Fig. 2 compatibility wall · clients not logos
FIG. 2

One memory server, every MCP client

One server. All your agents share the same memory. No lock-in.

Claude CodeCursorCodexGemini CLIany MCP client
Fig. 3 public signals · live when available
FIG. 3

Dev-native proof, without fabricated proof

The home page uses developer-native signals only: package download counts, registry status, and roadmap-labeled source availability. It does not invent customer quotes, logo walls, or measured result numbers.

Source availability

Client edges, managed engine

Public repo pending

The open-source split is documented without linking to a private or unpublished repository as public proof.

View open-source split
Registry listings

MCP Registry · Smithery · Glama

Launch listings pending

Registry links appear here only after each listing is accepted. The page does not present submission targets as live proof.

Install count

43 npm downloads last month

Fetched from npm for @answer-engine/mcp-server during the static build.

Fig. 4 four stages · ingest to inspect
FIG. 4

How the memory loop works

The product flow stays simple: bring sources in, normalize memory, recall it through MCP, and inspect the trail behind an answer.

4.1

Ingest

Bring in web pages, Atlassian content, CSVs, documents, and other source material without turning each agent into a bespoke ingest job.

IN: source material · OUT: normalized memory
4.2

Remember

Store tenant-scoped memory with summaries, tags, and lineage so agents can reuse context instead of starting cold in each tool.

SCOPE: tenant isolated · MODE: inspectable
4.3

Recall

Expose the same memory through MCP so Claude Code, Cursor, Codex, Gemini CLI, and any compatible client query one shared source of truth.

SURFACE: MCP · CLIENTS: any compatible agent
4.4

Inspect

Trace what was recalled, what source it came from, and what memory was superseded so the answer remains explainable.

TRAIL: source · lineage · supersession
Fig. 5 the wedge · inspector asset pending
FIG. 5

See what your agent remembered — and what it forgot

The wedge is not a performance claim. It is the inspectable cold path: source lineage, superseded memory, and missing-context evidence an operator can review when an answer needs proof.

Inspectable memory

Memory should be auditable, not mystical.

The inspector shows which sources were recalled, what memory was skipped, and where the system needs better source coverage. That lets teams improve memory without pretending a measured result exists.

View inspector roadmap
PLACEHOLDER — PRODUCT C.2

Inspector screenshot slot

Real screenshot pending the product inspector asset. This frame is intentionally labeled so it cannot read as shipped proof.

Fig. 6 security band · routed to trust detail
FIG. 6

Trust controls belong on the first page

Memory is sensitive infrastructure. The page states the controls that matter at the install decision and routes deeper evaluation to security.

hard tenant isolation no training on your data right-to-erasure
Fig. 7 ICP fork · figure cells
FIG. 7

Choose the path that matches your use case

The home page routes the developer who wants an install, the AI team evaluating shared memory, and the agency separating client contexts.

Fig. 8 open-source split · no engine overclaim
FIG. 8

Open where portability matters, managed where operations matter

The open-source promise is deliberately narrow: installable client surfaces stay inspectable and portable, while the hosted engine, index, and billing remain managed services.

Open

MCP server and CLI

The installable edge is open: @answer-engine/mcp-server and @answer-engine/cli. Read it, run it, and keep your agent client from being locked to a single app.

Open-source split
Managed

Hosted engine, index, and billing

The service that stores tenant-isolated memory, runs the hosted index, and handles account/billing operations is managed. This page does not imply the full engine is open source.

Read the split
Fig. 9 roadmap teaser · no vaporware as available
FIG. 9

Shipping next, labeled as next

These items are roadmap work, not available-now claims. Each one uses the Roadmap chip so the claim ledger remains visible in the page body.

Reranked retrieval Reproducible scorecard Org-shared mode Continuous-sync scheduler
FIG. 10 — START

Give your agent inspectable memory from your own sources.

Install the MCP server, connect a source library, and route every compatible agent to the same tenant-isolated memory layer.

Install: npx answer-engine-mcp
Package: @answer-engine/mcp-server
Clients: Claude Code · Cursor · Codex · Gemini CLI
Trust: no training on your data