What you get

MCP — the Model Context Protocol — is an open standard for connecting AI clients to outside tools and data. recuerd0 runs a remote MCP server at POST /mcp (Streamable HTTP, JSON-RPC 2.0): point an MCP client at your instance and it can read, search, and write memories across your workspaces, through six tools scoped to the access you approve.

The connection is consent-based. The client registers itself, you approve a short permission screen, and that's it — there is no API token to generate, copy, or paste into a config file. Approve once per client; disconnect any time.

Why MCP

MCP lets the model reach your curated context in the middle of a conversation, instead of you pasting it in.

Without it, you copy a personal API token into each tool's config and paste context into prompts by hand — one static token with broad access. With recuerd0 over MCP, you approve a consent screen once (the client registers itself), the model queries memories on demand — list, search, read — and access stays scoped and revocable: you can see every connected client and disconnect it.

Supported clients

Any client that speaks the Model Context Protocol over HTTP can connect.

ClientHow it connects
Claude DesktopAdd a custom connector with your server URL
Claude.aiAdd the connector in settings
Any MCP clientSend JSON-RPC 2.0 to POST /mcp (Streamable HTTP)

Connect Claude Desktop

Three steps. The OAuth handshake runs itself once you add the URL.

1. Open connector settings

In Claude Desktop or Claude.ai, go to Settings → Connectors → Add custom connector.

2. Add your server URL

Use the MCP endpoint of your instance.

server url
https://recuerd0.ai/mcp

3. Approve the consent screen

The first connection opens a browser screen titled Connect to Recuerd0. Approve the scopes and the client is connected.

You need to be signed in to recuerd0 in the same browser — the consent screen approves access on behalf of your account.

Authorization

recuerd0 is its own OAuth 2.1 authorization server — built into the app, no external service. Clients self-register with Dynamic Client Registration and prove themselves with PKCE; there are no shared secrets to manage.

EndpointPurpose
GET /.well-known/oauth-authorization-serverAuthorization server metadata (RFC 8414)
GET /.well-known/oauth-protected-resourceProtected resource metadata (RFC 9728)
GET /.well-known/oauth-protected-resource/mcpPath-suffixed variant clients probe first
POST /oauth/registerDynamic Client Registration (RFC 7591)
GET · POST /oauth/authorizeConsent screen and authorization code
POST /oauth/tokenExchange the code, or refresh an access token
POST /oauth/revokeRevoke a token (RFC 7009)

Check discovery

The authorization server metadata is public — a quick way to confirm an instance is reachable.

shell
curl https://recuerd0.ai/.well-known/oauth-authorization-server
Access tokens last one hour; refresh tokens rotate, so a connected client stays connected without asking you to approve again.

The six tools

The server exposes six tools. Three read, three write. Each one is scoped to your account's workspaces.

ToolTypeWhat it does
list_workspacesreadList the workspaces in your account
list_memoriesreadList memories in a workspace; optional FTS5 query and category filter
read_memoryreadRead the full content of one memory
create_memorywriteCreate a memory in a workspace, with optional tags
update_memorywriteUpdate a memory in place — title, content, category, or tags
create_versionwriteAppend a new immutable version to a memory, preserving history

A tool call

Tools are JSON-RPC 2.0 methods. The client handles this — you just ask the model in plain language.

json-rpc
{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "list_memories", "arguments": { "workspace_id": "1", "query": "authentication" } } }

Scopes & permissions

A client asks for scopes; recuerd0 maps them to a permission tier on the token it issues.

ScopeGrantsTier
memories:readRead and search memoriesread_only
memories:writeCreate and update memoriesfull_access
workspaces:readList your workspacesread_only

A read_only token can call list_workspaces, list_memories, and read_memory. The three write tools — create_memory, update_memory, and create_version — need full_access.

Every tool scopes to your account's workspaces. A token can never read or write another account's memories, whatever it asks for.

Connected applications

Every client you approve appears in your profile under Connected applications, with the name it registered and when it connected.

Disconnect a client

Disconnecting revokes the token immediately (RFC 7009). The client loses access until it reconnects and you approve it again.

Revoked tokens are kept as a record, not deleted — so a disconnect leaves an audit trail rather than vanishing.

Security & limits

The MCP and OAuth endpoints are rate-limited per token or per IP. Exceeding a limit returns 429 Too Many Requests.

EndpointLimit
POST /mcp120 requests / minute per token
POST /oauth/token30 requests / minute per IP
POST /oauth/register15 requests / minute per IP
POST /oauth/revoke30 requests / minute per IP

Clients prove themselves with PKCE rather than a shared secret, access tokens are short-lived, and tokens are stored only as hashes — the raw value exists just long enough to hand to the client.

Troubleshooting

SymptomLikely causeFix
Client can't authorizeThe instance URL isn't reachable, or the URL is wrongUse the full https://<instance>/mcp URL; confirm the instance is public
Consent screen sends you to sign inNo active recuerd0 session in that browserSign in to recuerd0 first, then retry the connection
Write tools are missingThe token is read_onlyReconnect and approve Create and update memories
App stopped working after a disconnectIts token was revokedReconnect — the client re-registers and re-consents