Developer setup

Saga is available as an MCP server, so AI agents and assistants — Claude, ChatGPT, Cursor, and custom agents — can use your Saga workspace as a tool. The agent can ask Saga questions, search and read project documents, and run your workflows, all using your own access. This is the reverse of connecting external MCP servers to Saga. Here, your Saga workspace is the server that other tools connect to.

Before you start

You need:

Your platform URL and an API key. Both come from Settings > API Key Management inside Saga. Open the page, create a key if you don’t have one, and copy both values.
An MCP client that supports the streamable HTTP transport (Claude Desktop, Claude Code, Cursor, or any custom agent built on the MCP SDK).

Each customer has their own Saga domain, so your MCP endpoint is your own platform URL with /api/mcp added. Use the platform URL shown in Settings > API Key Management — don’t guess the domain.

Connect a client

Most clients only need the endpoint URL and an Authorization header.

Endpoint: https://<your-domain>.sagalegal.io/api/mcp
Header: Authorization: Api-Key <your-api-key>

For a client that reads an .mcp.json file (for example Claude Code), add:

{
  "mcpServers": {
    "saga": {
      "type": "http",
      "url": "https://<your-domain>.sagalegal.io/api/mcp",
      "headers": { "Authorization": "Api-Key <your-api-key>" }
    }
  }
}

Replace <your-domain> with your platform URL and <your-api-key> with the key from Settings > API Key Management.

Some clients only let you paste a bearer token. In that case use Authorization: Bearer <your-api-key> instead — Saga accepts your API key in either form.

What the agent can do

Once connected, the agent has these tools:

Tool	What it does
`ask_saga`	Asks Saga a question and returns the answer, the same as chatting in the app.
`ask_saga_project`	Asks a question grounded in one project’s documents and returns a cited answer. The chat is saved in the project.
`list_projects`	Lists your projects, including ones shared with you.
`list_project_documents`	Lists the documents in a project.
`search_project_documents`	Searches a project’s knowledge base and returns the most relevant sections.
`fetch_project_document`	Returns the full text of a document.
`list_assistants`	Lists the active custom assistants in your workspace.
`ask_saga_assistant`	Asks one custom assistant a question, with its instructions applied.
`list_prompts`	Lists prompt-library entries (personal, organization, Saga Lab, or all), with optional search.
`fetch_prompt`	Returns one prompt’s text and its placeholder fields.
`list_workflows`	Lists your workflows and the inputs each one expects.
`execute_workflow`	Runs a workflow with text inputs and returns the result.
`get_workflow_status`	Returns a workflow run’s status and each step’s status.
`fetch_workflow_step_result`	Returns the output of one completed step.
`submit_workflow_input`	Answers the step a run waits at and continues the run.
`resume_workflow`	Starts a run that was created but never started.
`list_project_folders`	Lists the knowledge-base folders and tags of a project you can open.
`list_project_memories`	Lists the short facts saved in the memory of a project you can open.
`list_chats`	Lists the personal chats and project chats you can open in the app. Chats past the retention window are left out.
`fetch_chat`	Returns the messages of one personal or project chat you can open. Long chats are truncated to the token budget.
`list_grid_reviews`	Lists your grid reviews. Pass `project_id` to list only the reviews of one project.
`fetch_grid_review_answers`	Returns the answers of one grid review you can open. Large reviews are truncated to the token budget.
`search_legal_research`	Searches one legal database that you have access to. A jurisdiction outside your workspace is refused.
`list_workflow_runs`	Lists the workflow runs that you started. Use `get_workflow_status` to read one run.
`fetch_workflow`	Returns the steps of a workflow in order. For each input step, it shows what you must supply and whether the agent can supply it over MCP.
`describe_action`	Explains how to use the Saga MCP tools. Without arguments, it lists every tool and the usual call order. With a tool name, it returns the inputs, an example call, what the tool returns, and what to call next.

Every tool runs as you. The agent sees only the projects, documents, chats, grid reviews, and workflows your account can already access — the same permissions you have in the web app.

Security

Treat your API key like a password. Anyone with the key and your platform URL can act as you through Saga.
Rotate or revoke the key at any time from Settings > API Key Management. Revoking it immediately stops any connected agent.
If a key may have leaked, revoke it and create a new one, then update your client configuration.

Troubleshooting

The client can’t connect or returns 401 Unauthorized — check that the endpoint ends in /api/mcp, that the header is Authorization: Api-Key <your-api-key>, and that the key hasn’t expired or been revoked. Create a fresh key in Settings > API Key Management if needed.
A workflow waits for input: use submit_workflow_input to answer a text input and continue the run. If the step needs a file, open the run in Saga and upload the file there.

Welcome

Core capabilities

Integrations

Account types

Administration

Troubleshooting

Developer setup

Before you start

Connect a client

What the agent can do

Security

Troubleshooting

​Before you start

​Connect a client

​What the agent can do

​Security

​Troubleshooting

Before you start

Connect a client

What the agent can do

Security

Troubleshooting