AGENT CONTEXT / DOCS

Connect your coding agents.

Use the same project ID and database across clients.

Choose your MCP client

Any coding agent with MCP tool support can connect over stdio or Streamable HTTP, regardless of the AI model it uses. Agents without MCP support need an MCP integration.

Merge into your client’s MCP configuration. Adapt the wrapper if required by that client. Merge with existing settings rather than replacing the file.

agent-context-mcp config --client generic --db /absolute/path/to/shared/context.sqlite3

{
  "mcpServers": {
    "agent-context": {
      "command": "agent-context-mcp",
      "args": [
        "stdio",
        "--db",
        "/absolute/path/to/shared/context.sqlite3"
      ]
    }
  }
}

Use the same absolute database path in every local agent and the dashboard. Install the package first, enable MCP tools in your client, then start a fresh agent session. If the executable cannot be found, use its absolute path in the command field.

For a client on another computer, connect to the server’s reachable /mcp URL with a bearer token and Streamable HTTP support. Local stdio paths and localhost refer to the computer running the client.

Using a local model or another operating system? Read the compatibility guide. Connecting teammates on separate computers? Follow the LAN, VPN, SSH, or HTTPS setup guide.

Local agents over stdio

Each agent starts its own process and shares the configured SQLite file. The explicit stdio action is required because the default npm command starts the dashboard.

codex mcp add agent-context -- agent-context-mcp stdio --db /absolute/path/to/shared/context.sqlite3
claude mcp add --scope user agent-context -- agent-context-mcp stdio --db /absolute/path/to/shared/context.sqlite3

Agents over HTTP

The following commands connect a client on the same computer as the server. For a teammate’s computer, use the endpoint and host token from the team guide. Do not generate a separate local token on each teammate’s computer.

Keep the server running. Export its token in the terminal that launches your client:

export AGENT_CONTEXT_TOKEN="$(agent-context-mcp token)"
codex mcp add agent-context --url http://127.0.0.1:8765/mcp --bearer-token-env-var AGENT_CONTEXT_TOKEN

For Claude Code:

claude mcp add --scope user --transport http agent-context http://127.0.0.1:8765/mcp --header "Authorization: Bearer $AGENT_CONTEXT_TOKEN"

Claude saves the resolved token in its configuration. Protect that file. On PowerShell, use $env:AGENT_CONTEXT_TOKEN = agent-context-mcp token and the same environment-variable syntax in the header.

Make the shared context part of the workflow

Add this instruction to the file your agent reads in the repository you are building, such as AGENTS.md, CLAUDE.md, GEMINI.md, or editor rules:

Use agent-context MCP for project ID "my-project".
Read get_context and list_tasks before working.
Use a unique agent_id and claim your task.
Record progress, changed files, decisions, and test results.
Before stopping, write a handoff and complete or release your task.

Use a different agent ID per session, and the same project ID in every client. Restart existing agent sessions after changing MCP registration.