Claude
Use a configured Claude provider in Astir, or connect Claude Code, Desktop or claude.ai directly to Becki.
03 / Connect your AI
One account.
A working connection.
Install Becki and connect each client you use. Then check that the agent can retrieve your context and save new work. Installing the server alone does not do that.
Use Core or Astir to work with Becki. For an existing client, jump to its connection steps. Already connected? Add the memory protocol and verify with real context.
01 / Two routes
A model in Astir.
Or memory in your client.
OpenAI
Use an OpenAI model in Astir, or connect Becki to Codex or ChatGPT using the client instructions below.
Grok / xAI
Use a Grok model in Astir, or connect Grok.com directly to Becki.
Provider credentials and model access depend on your setup. Astir's Connections manages providers, tools and MCP servers inside the workspace. External AI clients need their own Becki connection.
02 / Install
Give Becki
a place to run.
Core / Windows, macOS, Linux
The $15/mo memory daemon uses two npm packages: becki-mcp is the MCP server; becki is the terminal admin interface. No meeting capture or native Mac workspace.
- Install the packages, then run
becki-mcp initto register projects and print your MCP configuration. - Run
beckiand sign in with your Becki account in the browser when prompted. - Add the MCP snippet printed by initialization to your client's configuration. Use your own generated configuration rather than guessing an application path.
- Use the
beckiadmin interface for watched folders, backfills and subscription management.
npm install -g becki-mcp becki
becki-mcp init # registers projects, prints MCP config snippet
becki # admin TUI: browser sign-in, run backfillsOptional history backfill: this command walks 90 days of Claude Code, Cursor and Codex session logs. Review which work you are authorizing before running it.
becki-mcp bootstrap 90Astir / Mac with Apple silicon
Astir is $25/mo or $250/yr, with a 14-day free trial, and requires a Becki login. It brings your agent workspace and Becki's memory together. Use the installer requirements for the release you download.
- Sign up for Astir.
- Get the current Astir Mac installer from the download portal and follow its installation steps.
- Sign in with the Becki account you use for your memory.
- Review Connections for the providers, tools and MCP servers you want to enable.
- Review microphone, calendar and screen/audio-capture permissions when using those features. You can decline a permission; installing the app does not grant it.
Keep your existing Becki account and vault. Configuring Astir is separate from registering Becki in another editor or chat client.
Review the Astir plan03 / Connect per client
Pick the client
you actually open.
Claude Desktop, Claude Code and claude.ai
Core users use the snippet printed by becki-mcp init. A connection configured inside Astir does not automatically register these external clients.
- Open the client's MCP configuration and use the entry generated for your Core installation, or its supported authenticated remote connector.
- Restart the client if it loads configuration at startup.
- Authenticate with your Becki account when requested.
- Confirm that the client lists the Becki server and tools, then ask the verification question below.
claude.ai in the browser
- Open Settings → Connectors → Add custom connector.
- Paste the endpoint below.
- Use OAuth to sign in with the same Becki account you use for Core or Astir and review consent.
- Use the connector in a conversation and verify that it returns your actual context.
https://mcp.becki.io/mcpYou can also use the connector in the claude.ai phone app. Use your authenticated account and confirm the connector is available in the session; this is not a mobile recording app.
Core users: use your initialization output for local stdio setup. Claude Code can also use the skills pack in the protocol section.
Cursor
- Open Cursor's MCP configuration and add the local Core snippet printed by
becki-mcp init, or use the remote MCP endpoint below. - For the remote connection, use OAuth to sign in with your Becki account and review consent.
- Restart Cursor if needed, then confirm Becki's tools are listed.
- Ask the verification question against your project context and add the memory protocol to Cursor Rules.
https://mcp.becki.io/mcpCodex CLI, IDE extension and desktop
- Use the configuration snippet from
becki-mcp initfor local Core setup, or Codex's supported remote MCP configuration with your Becki account. - Restart the Codex client or IDE after changing its MCP configuration.
- Confirm Becki's tools appear, then ask what it knows about your current project.
- Add the protocol or install Becki's skills so future sessions retrieve and capture meaningful context.
These steps cover Codex CLI, the VS Code / JetBrains extension and Codex desktop. The cloud sandbox at chatgpt.com/codex cannot reach a local MCP server. Do not treat local registration as a cloud-sandbox connection.
ChatGPT's browser connector is a separate route, with its own developer-mode setup in the ChatGPT tab.
Grok.com
Grok.com connects directly to Becki over MCP. You do not need Astir for this connection.
- Add a remote MCP connection in Grok.com using the endpoint below.
- Select OAuth authentication and sign in with your Becki account.
- Review consent and authorize the Becki connection.
- Enable Becki in your conversation and ask the verification question to check your actual project context.
https://mcp.becki.io/mcpUsing a Grok model through an xAI provider in Astir is a separate option. It does not configure your Grok.com connection.
ChatGPT in the browser
- Turn on developer mode in your ChatGPT account.
- Add a new connector and use Becki's remote MCP endpoint below.
- Use OAuth to sign in with your Becki account and review the consent screen.
- Enable the connector for the conversation and ask the verification question.
https://mcp.becki.io/mcpAvailability depends on the ChatGPT account's connector and developer-mode support. These steps are for web ChatGPT, not a promise of ChatGPT Desktop or Codex cloud access.
Perplexity
- Check your plan. Custom connectors require Perplexity Pro, Max or Enterprise, not the free tier.
- Open Settings → Connectors → Add custom connector.
- Use the remote MCP endpoint below and sign in with your Becki account through OAuth.
- Enable the connector and verify a question against your own project history.
https://mcp.becki.io/mcpA paid client plan is separate from your Becki subscription. It does not grant access to another person's vault.
Zed
- Open Zed's MCP configuration and add the local Core snippet printed by
becki-mcp init, or its supported authenticated remote connector. - For a remote connection, use
https://mcp.becki.io/mcpand authenticate with your Becki account through OAuth. - Restart Zed if it loads configuration at startup, then confirm Becki's tools appear.
- Add the memory protocol to Zed AI settings and ask the verification question.
Becki's VS Code sidebar extension
VS Code can be an MCP client. Becki also has a separate Marketplace extension for memory in the sidebar.
- Open Extensions and search for Becki, or use the command below.
- Use the extension's Connect action on first launch. It talks to Becki's remote server.
- Check Briefing, This Project, Commitments and Ask. If they show unexpected zeros, first confirm you are signed in.
code --install-extension becki.becki-vscodeInstalling the sidebar extension is not the same as configuring MCP for another agent inside the editor. Use that agent's own connection setup when required.
Raycast
- Open Raycast's MCP configuration and add the Core snippet printed by
becki-mcp init, or its supported authenticated remote connector. - For a remote connection, use
https://mcp.becki.io/mcpand sign in with your Becki account through OAuth. - Reload the client if needed, then confirm it lists the Becki server and tools.
- Ask the verification question and add the memory protocol to the agent's instructions.
Antigravity CLI / agy
Antigravity has its own MCP configuration keys and files.
- Register the server in
~/.gemini/config/mcp_config.json. The key isserverUrl, noturlorhttpUrl. Merge this entry into your existing file rather than overwriting unrelated servers.
{
"mcpServers": {
"becki": {
"serverUrl": "https://mcp.becki.io/mcp"
}
}
}2. If you want Becki tools available in background sessions, review this permission rule for ~/.gemini/antigravity-cli/settings.json. It allows all tools on the Becki server without per-call prompts; enable it only if that scope matches your authorization.
{
"permissions": {
"allow": ["mcp(becki/*)"]
}
}3. Authorize Becki in each interface you use. The desktop and CLI read the same MCP registration but retain separate sign-in tokens. Review each configuration; connecting Astir does not grant access in Antigravity.
04 / Use the memory
Connected is not
the same as remembering.
Install the workflows
Astir can use skills in its agent workspace. For external clients with Core, install the Becki plain-text skills pack with the command below. No sign-in is needed to install the pack.
becki skills installThe pack includes using-becki and becki-voice, plus catch-up, capture, ledger reconcile, rejected options, handoffs, evidence tracing, and source-aware code work. using-becki requires becki_ingest in the same turn a durable decision, commitment, directive, priority shift, dead end, or explicit remember happens, in chat or on a live voice call. Chats and voice calls end before a closing turn runs. becki-voice is the writing-voice skill in that same pack. Skills call Becki's MCP tools. They do not create a second memory store.
Read the pack as plain text, including using-becki and becki-voice: becki-capture · becki-catchup · becki-reconcile · becki-decide · becki-handoff · becki-why · becki-code.
Keep your edited skills. With Core, use --force only if you intend to replace your edits with the shipped version.
Or add an explicit protocol
Read the standing authorization beside this before adding the protocol to your agent rules, such as Cursor Rules or Zed AI settings. It authorizes storing work context, not secrets.
For Claude Code, use project CLAUDE.md or global ~/.claude/CLAUDE.md. Review the standing authorization before adding it. Installing a skill does not itself grant permission to ingest private work.
Use the `becki` MCP tools as your shared memory.
Standing authorization from me, the user: I authorize my AI tools to
store my project and work context in my Becki vault. My vault is my own
account. This authorization never extends to secrets, credentials, or
keys, which must not be stored anywhere.
- Query `becki_context` before any non-trivial task.
- Call `becki_ingest` in the same turn a durable decision, commitment,
directive, priority shift, dead end, or explicit "remember" happens,
in chat or on a live voice call. Include the WHY. Chats and voice
calls end abruptly, before a closing turn runs.
- When the user references prior context, query Becki before asking
them to re-explain.
- Remember/forget instructions take effect in that same turn.Resolve a tracked task when it is done. If a decision changes, save the new reasoning and link it to the decision it replaces. That keeps completed work and old choices from appearing as current.
05 / Verify & troubleshoot
Ask for evidence.
Not a connection badge.
Open a connected client and ask the question below. Check that the answer uses real context from your work. If your vault is empty, you may need to authorize a project backfill first.
What does Becki know about me?The agent says there is no Becki tool.
In Astir, review the enabled Becki MCP connection in Connections. In an external client, review that client's MCP configuration and authentication, then restart it if required. In Core, check the snippet from initialization. A working server in one client does not mean every client has been configured.
The client is connected, but the vault looks empty.
First check that you are signed into the intended Becki account. In the VS Code extension, use Connect. Then check whether the project has been ingested and whether the agent has been capturing meaningful context.
Review the projects and history you want to ingest before authorizing a backfill. A connection badge alone does not establish that project history exists in the vault.
Queries return weak or incomplete context.
Confirm the relevant project is enabled and present in the vault. Ask about a specific decision rather than a broad summary. Compare the result with in-app Ask Becki, which uses hybrid retrieval; some MCP clients trim results. Install skills or rules so the agent retrieves before work and calls becki_ingest in the same turn, in chat or on a live voice call.
Meeting transcripts are blank or CPU stays busy.
Astir needs the relevant microphone and audio-capture permissions. Without audio capture, the remote side of a call may be missing. Review capture and transcription status in Meetings. Meeting capture is not a Core feature.
Memory contains an old decision and its replacement.
Becki's memory review proposes possible supersessions; it does not silently apply them. Ask a connected agent to review proposals. Approving ranks the older entry below the newer one without deleting history. Rejecting leaves both in place. Unreviewed proposals wait.
06 / Know the data path
Local capture.
Connected retrieval.
Local transcription processes audio on your device; configured cloud transcription sends audio to the selected provider. Synced vault text supports remote MCP retrieval, and selected context can be processed by configured AI providers. Becki is not zero-knowledge. Review the chosen feature, provider terms and account settings before sharing sensitive work.
Read Trust, Security and Privacy before authorizing ingestion or broad tool permissions. Do not place API keys, passwords or install tokens in memory prompts.
For export, account controls or current client support, contact support@becki.io.