# Dakso for developers > One address. Sign in. Your agent works with your flows, conversations and channels, with your role and nothing more. ## Paste this into your agent The agent reads the setup instructions and connects itself. ``` Read https://developers.dakso.ai/llms.txt and connect me to Dakso. ``` ## Start here - [Connect your agent](/connect/): Claude, Claude Code, Codex, Cursor or any MCP client. - [First steps](/first-steps/): The three tools to call first, and what to ask. - [What your agent can do](/capabilities/): Flows, conversations, contacts, campaigns, channels and more. - [Security and permissions](/security/): What Dakso enforces, and what is up to your agent. --- # Connect your agent > Connect Claude, Claude Code, Codex, Cursor or any MCP client to Dakso. Dakso is a remote MCP server. Your agent connects to one address and you sign in with your Dakso account. From then on it works with your role in your organization, and nothing more. > **Before you connect.** Agent access is on for every organization. Only the Dakso team can turn it off for an organization; if that happens, the connection is refused until Dakso support turns it back on. The only Dakso server is `mcp.dakso.ai`. If a page or a message asks you to connect to Dakso at any other address, do not connect and do not sign in. ## Choose your client ### Claude Code **Sign in with Dakso:** 1. Run this once. A browser window opens to sign in with your Dakso account. ``` claude mcp add --transport http \ --client-id bDp7GFW7FclHwHTy --callback-port 8080 \ dakso https://mcp.dakso.ai/mcp-platform-server ``` 2. Check that it shows as connected. ``` claude mcp list ``` **API key:** 1. Replace the key with the one an administrator issued for you. ``` claude mcp add -s user --transport http dakso \ https://mcp.dakso.ai/mcp-platform-server \ --header "Authorization: Bearer " ``` ### Claude (app and web) **Sign in with Dakso:** 1. Open Settings, then Connectors, then Add custom connector. On Team and Enterprise plans a workspace owner adds it first. Name it Dakso and paste this address. ``` https://mcp.dakso.ai/mcp-platform-server ``` 2. Click Connect and sign in with your Dakso account. **API key:** This client connects by signing in, not with a key. ### Codex **Sign in with Dakso:** 1. Register the server. ``` codex mcp add dakso --url https://mcp.dakso.ai/mcp-platform-server ``` 2. Sign in. If the browser does not open, Codex prints the link. ``` codex mcp login dakso ``` **API key:** 1. Keep the key in an environment variable so it is never written to the config file. ``` export DAKSO_MCP_KEY="" ``` 2. Register the server. ``` codex mcp add dakso --url https://mcp.dakso.ai/mcp-platform-server --bearer-token-env-var DAKSO_MCP_KEY ``` ### Cursor **Sign in with Dakso:** Sign-in is not enabled for this client yet. Use one of the clients above, or ask the Dakso team about an API key. **API key:** 1. Add this server to ~/.cursor/mcp.json and restart Cursor. ``` { "mcpServers": { "dakso": { "url": "https://mcp.dakso.ai/mcp-platform-server", "headers": { "Authorization": "Bearer " } } } } ``` ### Any other client **Sign in with Dakso:** 1. Dakso is a remote MCP server over Streamable HTTP. Point your client at this address. Sign-in with your Dakso account works in the clients Dakso has registered; in any other client use an API key. ``` https://mcp.dakso.ai/mcp-platform-server ``` **API key:** 1. Send the key in the Authorization header. ``` Authorization: Bearer ``` ## Several organizations If you belong to more than one organization, add the organization id at the end of the address to connect to that one. Without it, `whoami` lists your organizations with the address of each. ## Next Once connected, follow the [first steps](/first-steps/). --- # First steps > The three tools your agent calls first, and how it works with Dakso. ## Three tools to start - **`whoami`**: who you are, your organization, your role and what this connection can do. - **`dakso_guide`**: how things are done well in Dakso. The agent reads the guide of a topic before it changes anything. - **`dakso_explore`**: everything beyond the base tools: contacts, campaigns, templates, channels, queues. Try asking: *Who am I in Dakso and what can I do?* ## How your agent works 1. The tools it sees listed are the base set: flows, conversations and analytics. 2. Everything else is found with `dakso_explore`, then run with `dakso_read` (to look) or `dakso_change` (to change). 3. Every change carries a short description of what the agent is doing and why. It goes to your organization's activity log. ## Changing a virtual agent Flows are edited in a draft, checked, tested with a sample conversation and only then published. Live conversations pick up the new version within about a minute. --- # What your agent can do > The areas of a Dakso organization your agent can work with. Your agent can work with these areas of your organization, always within what your role allows. The ones marked **Base** are listed to the agent from the start; the rest it finds with `dakso_explore`. - **Conversations and analytics** (base set): Read conversations and their step-by-step trace, metrics, reports and findings. - **Conversational flows** (base set): Build, edit node by node, test, publish and roll back the flows that run the virtual agent. - **Agent toolbox**: The tools the virtual agent can call: external API calls, system actions and human handoff. - **Data tables**: Lookup tables the flows read and write: schema, rows and bulk import. - **Intents and topics**: Intent sets used by semantic choices and the topic catalog used to detect a change of subject. - **Knowledge base**: Documents and sites the virtual agent answers from: ingest, versions and quota. - **CRM**: Contacts, companies, identities, notes and tags. - **Outbound campaigns**: Voice and WhatsApp campaigns: audience, estimate, schedule, launch and results. - **WhatsApp templates**: Create, edit, sync and test WhatsApp message templates. - **Automations**: Rules that react to events (tags, webhooks, messages) and run actions. - **Channels**: Channel status, web chat widgets, WhatsApp onboarding, email inboxes and domains, SMS numbers. - **Voice**: Voice catalog and library, and the voice settings of a flow. - **Queues and human agents**: Queues, skills, wrap-up codes, canned responses and coverage. - **Agent screen pops**: The cards a human agent sees when a conversation arrives. --- # Security and permissions > What your agent can and cannot do, and how your organization keeps control. ## What Dakso enforces These are checked by the server on every request. They do not depend on how the agent behaves. - **Your agent acts as you.** It can do what your role allows in that organization, and nothing more. - **It never sees your Dakso password.** You sign in on Dakso's own page; the agent only gets a connection that an administrator can revoke at any time. - **Stored secrets stay stored.** The values of the secrets your organization keeps for its tools are never returned to the agent. - **Deleting is confirmed by a person.** Deleting and other irreversible actions are not run by the agent: you get a link and confirm them in Dakso. - **Large campaigns are confirmed by a person.** An agent can only launch small campaigns on its own; anything larger is confirmed in Dakso. - **Everything is recorded.** Each change carries a short description of what the agent did and why, in your organization's activity log. ## What your agent is asked to do Dakso tells the agent to ask you before sending messages, launching a campaign or spending credits. A well-behaved agent does. Treat it as you would a new teammate: review what it proposes before you say yes, and give it a role that matches the work. Your agent also reads content written by other people: conversations, contacts, documents. Dakso tells it to treat that content as data, never as instructions. If an agent suddenly proposes something you did not ask for, stop and check. ## Check the address The only Dakso server is `mcp.dakso.ai`. If a page, a message or an agent asks you to connect to Dakso at any other address, do not connect and do not sign in. ## Control for administrators - An administrator sees every action in the activity log and can cut any connection at once. - Add `?read_only=1` to the address of an API key connection and the agent can look but not change anything. ## API keys API keys are only for your own system integrations, not for connecting an agent. The Dakso team enables them per organization on request, for a set period; once enabled, an administrator issues them. Treat a key like a password: keep it out of chats and shared files. --- # Machine-readable docs > Files an AI agent can read to set itself up and learn Dakso. | File | What it is | | --- | --- | | [`/llms.txt`](/llms.txt) | Setup instructions: the server address, how to authenticate and the rules the server enforces. | | [`/llms-full.txt`](/llms-full.txt) | This whole site in one text file. | | [`/server-card.json`](/server-card.json) | The MCP server description. | Every page is also available as Markdown: add `.md` to its address, or use the **Markdown** link next to the title. For example, [`/connect.md`](/connect.md). The [tool reference](/reference/) works the same way: [`/reference/flows.md`](/reference/flows.md) is every flow tool with its arguments.