Skip to content
MCP server

Connect your AI to Talavine.

Connect your email, files, notes, and calendars to Talavine once. Then connect a compatible AI app to Talavine to reach the information it needs. Claude Desktop, Cursor, and Codex have one-click setup inside Talavine; other compatible apps use a short configuration entry. Choose another AI for a different job without reconnecting each data source. You control which apps connect and what they can access.

What it is

One MCP. Every provider. Source-grounded.

An AI app's own connectors usually reach one service at a time. Talavine gives it a single MCP server across every account you connect, with thread context, person history, commitments, and extracted facts ready to use.

Runs locally, stdio by default

The server is Talavine.McpServer, a small .NET process speaking MCP over stdio. Logging goes to stderr; JSON-RPC owns stdin/stdout. An optional HTTP mode listens on this computer only (loopback) and requires a bearer token. No remote service to trust.

Reads the same data the app reads

Points at the Talavine data directory (env var Talavine_DATA_DIR if you want to override). Per-account SQLite databases, knowledge-claim store, context-fact projections. The MCP server is a read & write surface over the same files the desktop app uses.

Tools, not endpoints

Dozens of typed MCP tools, organized by capability. Read tools are marked ReadOnly so agents that respect MCP semantics can run them without confirmation. Mutations land in a durable outbox and replicate to the provider on the next sync.

Tool catalog

What your agent can actually do.

The catalog is grouped by intent. Tools return JSON strings the model can consume directly; most read tools accept account scoping when you need it and span every account when you don't.

Context: the flagship

ToolWhat it returns
get_context_packQuery → source-backed facts with confidence scores and citations, redacted for outside clients. A client you granted context.sensitive can ask for the fuller trusted-peer view.
get_thread_contextThread → participants, action items, commitments, and a last-N message rollup. The thread is JWZ-stitched across References / In-Reply-To, not just subject-matched.
get_person_contextPerson → relationship history across every connected account at once. Common topics, last interaction, response patterns.
get_company_contextCompany / domain → who you talk to there, what about, and what's currently in flight.

Search & read

ToolWhat it returns
search_messagesFull-text search across every account in one call. Returns matches with thread, account, and source-ref.
search_messages_semanticVector search over enriched message embeddings. Use when the caller doesn't remember the words.
find_related_messagesGiven a message, find its semantic neighbors. Good for "what else like this?" follow-ups.
search_entities · search_contacts · search_contacts_semanticKnowledge-graph entity lookup. People, companies, projects, all extracted from real communication, not a directory import.
list_recent_messages · get_message · get_thread · list_accounts · unread_countThe everyday read primitives. Account-scoped where it matters; cross-account where it helps.

Act on the inbox

ToolWhat it does
send_message · reply_to_email · forward_email · save_draftCompose and send across the right provider. Native Gmail / Microsoft Graph paths used when available; SMTP only as the floor.
mark_message_read · flag_message · delete_message · move_message · snooze_messageMutations are durable. They land in the outbox first and replicate to the provider on the next sync, surviving crashes, restarts, and offline windows.
list_folders · list_rules · create_rule · toggle_rule · delete_ruleInspect and shape the inbox policy the agent will see. Rules are user-authored plain English.

Talavine agent operations

ToolWhat it does
list_agent_profiles · apply_agent_role · request_agent_cycleInspect Talavine's internal agent profiles, swap roles, request a cycle. Useful when you're orchestrating Talavine from a larger agent system.
get_recent_journal · list_agent_proposalsRead what Talavine's own agents have been doing and what they want to propose. Audit trail in, audit trail out.
Wire it up

One click for Claude Desktop, Cursor, and Codex.

In Talavine, open Settings → MCP Connectors and click Connect next to the app under Use your data hub in other AI apps. Talavine adds itself to that app's settings; quit and reopen the app and it can search and read the profile you have open. That connection can't send email or change anything in Talavine, and you can disconnect at any time.

Claude Code and other MCP clients need an entry that launches the MCP server installed with Talavine. The easiest way to get it right is to copy it: Settings → MCP Connectors → Connect outside agents has ready-made Claude Code and Codex entries with this computer's exact path, your active profile (--data-dir), and the access you choose. The examples below show the shape.

The server is /Applications/Talavine.app/Contents/MacOS/Talavine.McpServer on a Mac. On Windows it's Talavine.McpServer.exe beside Talavine.App.exe in your install folder, for example C:/Users/you/AppData/Local/Programs/Talavine/Talavine.McpServer.exe for a per-user install.

Claude Desktop

Edit claude_desktop_config.json (on macOS at ~/Library/Application Support/Claude/, on Windows at %APPDATA%\Claude\), then fully restart Claude Desktop:

{
  "mcpServers": {
    "talavine": {
      "command": "/Applications/Talavine.app/Contents/MacOS/Talavine.McpServer",
      "args": []
    }
  }
}

Claude Code

Run claude mcp add, or put the same entry in a project's .mcp.json:

claude mcp add talavine -- "/Applications/Talavine.app/Contents/MacOS/Talavine.McpServer"

Codex CLI

Codex reads MCP servers from ~/.codex/config.toml:

[mcp_servers.talavine]
command = "/Applications/Talavine.app/Contents/MacOS/Talavine.McpServer"
args = []

Any MCP-aware client

The transport is plain MCP stdio; any client that speaks the protocol can launch the binary. The server identifies itself as Talavine v1.0.0 on handshake and advertises the full tool catalog from WithToolsFromAssembly().

Access and redaction

Each client gets the access you grant it.

You choose a client's access when you connect it, and the MCP server enforces those scopes. A client can't ask for more by itself: the view a context tool returns follows the client's grant, not a level the caller picks.

External view

What every MCP client gets. Personal identifiers and internal IDs are redacted, and only fields approved for external use come back. Exporting context at all needs the context.export scope.

Trusted-peer view

Includes more sensitive local facts. A client gets it only if you granted the separate context.sensitive scope and it asks for it.

Internal view

For Talavine's own agents on your computer. MCP clients can't request it.

Trust model

The data layer enforces what the prompt can't.

An MCP server is only as safe as what flows back through it. These data-layer guards limit what an outside agent can see and what a message can change.

  1. Knowledge promotion is gated A single inbound message can create a KnowledgeClaim but never a durable UserMemory. Promotion requires user-authored, trusted-contact, or corroborated status. Inbound content is fenced, and a tool-boundary guard can block or escalate consequential calls that appear authored by fenced content.
  2. Sources travel with facts Every durable record carries a SourceRef: kind, account ID, record ID, excerpt, observed-at. An agent that returns a fact can always be asked to surface the message. No hidden provenance.
  3. Deletion cascades or tombstones Delete a message and registered IMessageDeletionObserver implementations dispatch cleanup. Derived data either disappears or tombstones with a recorded source-was-deleted reason. Stale data doesn't accumulate.
  4. Suppression survives reset If you reject a claim, a KnowledgeSuppression row blocks the system from relearning it from a future message. The agent can't talk you back into it by quoting a duplicate.
  5. Routing respects privacy classes LLM calls Talavine makes on the agent's behalf go through ILlmRouter. Set PrivacyClass = LocalOnly on sensitive function types and prompts never leave the machine, enforced before any cost / quality advisor runs.
Plug it in

Download Talavine, connect your accounts, point your agent at the MCP.

Your agent now reads, queries, and acts across every inbox you own, with citations, redaction, and a memory layer that won't quietly drift.