Overview
Atlas supports the Model Context Protocol (MCP), an open standard that lets compatible assistants work with external systems. In Atlas, that means you can ask an assistant to find civic actors, check source trails, compare place coverage, save useful records, and start new source-linked research from the conversation. The assistant performs the MCP operations. You stay in control of the choices that affect your workspace, budget, account, or saved research. Example conversation with Claude:You: Who’s working on housing issues in Kansas City? Claude: I found 23 entities working on housing in Kansas City. Here are the most prominent… You: What are the coverage gaps? Claude: Kansas City has strong coverage on housing affordability and tenant organizing, but limited coverage on homelessness services and community land trusts…
What you need
- An Atlas account — Sign up if you haven’t already.
- Atlas Pro, Team, or Research Pass — MCP access requires a paid plan (see pricing). Free-tier sign-ins can complete, but Atlas assistant capabilities stay unavailable until a plan is active.
- An MCP-compatible client — Claude Desktop, Claude Code, VS Code with Copilot, Gemini CLI, Codex CLI, or Copilot CLI.
Before you connect
- Bootstrap allowlist. Atlas runs an operator allowlist while we’re inviting users from the bootstrap list. If your email isn’t on it and you don’t have a workspace invitation, the OAuth sign-in won’t deliver a magic link. Request access at atlas.rebuildingus.org before connecting an MCP client.
- Workspace binding. After signing in, your primary workspace is selected automatically. Multi-workspace users see a workspace picker on the consent screen.
- Session lifecycle. Most MCP clients keep the connection fresh automatically. If Atlas asks you to sign in again, complete the browser flow.
What Atlas can help with
After you connect Atlas, you can ask an assistant to:- Find people, organizations, initiatives, campaigns, and events by place, issue area, source type, or search text.
- Show where a claim came from by opening the sources behind an actor or research run.
- Compare civic coverage across places and issue areas, including where Atlas has strong evidence and where coverage is thin.
- Save selected actors to private workspace lists.
- Create private coverage targets that help your workspace track a place, issue, actor type, or evidence gap.
- Watch actors or coverage targets for source-backed changes.
- Start a new research run when existing Atlas coverage is not enough.
- Open Atlas account settings from the conversation when you need to manage billing or API-key setup.
Connection details
Atlas runs a remote MCP server at:Available capabilities
Once connected, your assistant can use these Atlas capabilities:Available resources
Atlas resources let compatible clients reopen selected research outputs:resources/list returns only a short set of recent research runs and Atlas UI
resources. Ask your assistant to find the relevant records first, then use
returned resource_uri values to open the same records later.
Guided research starters
Some MCP clients can show Atlas research starters as selectable commands. Use them when you want a guided starting point for common civic research jobs: researching a place, finding civic actors, inspecting an entity’s source trail, assessing coverage gaps, or creating a brief from an existing discovery run. Research starters are read-only in this version. They help the assistant follow the right Atlas source-checking steps, but they do not create discovery runs or modify workspace data.Clarification and confirmations
Atlas may ask for a small clarification or confirmation before the assistant continues. Read-only requests use this only when a choice materially changes the result, such as an ambiguous place or issue-area match. Workspace actions require explicit confirmation before saving actors, creating coverage targets, watching resources, or spending a research run. Sensitive interactions stay on Atlas-owned browser pages. Billing settings use browser consent instead of passing payment information through the MCP client. See Confirmations and follow-up questions for the full privacy and troubleshooting boundary.Long-running research runs
Most Atlas answers come from data that already exists. New research runs create fresh source-linked research, so compatible clients receive a task handle and can check progress while the run finishes. Clients without task support receive an explicit error. Uselist_discovery_runs or get_discovery_run to check on runs started elsewhere.
Workspace admin activity
Workspace admins can see successful integration activity by surface: total calls, API calls, MCP calls, top paths, and last-seen timestamps. Atlas does not expose request payloads, prompts, raw metadata, private notes, or session replay in that view. See Workspace integration activity for the access requirements and data boundary.Connect from a compliant client
Every supported client uses the same Atlas endpoint. Use your MCP client’s built-in flow for adding a remote MCP server, then provide the Atlas server URL:Troubleshooting
“Connection failed” or timeout Check thathttps://atlas.rebuildingus.org/mcp is reachable from your network. If you’re behind a corporate proxy, ensure it allows outbound HTTPS.
OAuth window doesn’t open
Make sure your default browser is set correctly and your MCP client supports remote OAuth login.
“Unauthorized” after connecting
Your OAuth token may have expired. Disconnect and reconnect the server in your client to trigger a fresh authorization flow.
Atlas actions don’t appear
Restart your MCP client after connecting. Some clients require a reload to discover newly available Atlas actions.
For more help, see MCP troubleshooting or the general Troubleshooting guide.
References
Atlas publishes the hosted server URL, required plan, available capabilities, scopes, and Atlas-side troubleshooting. Client-specific menus and command names belong in the client documentation.- MCP Registry — Atlas’s prepared registry namespace and metadata status.
- MCP resources — Research outputs that can be opened again.
- Authentication reference — Token lifecycle, refresh rotation, and discovery URL catalog.