The MCP server for Claude Code that runs your backlog from the terminal.

Claude Code has deep filesystem and terminal capabilities, but every session starts from zero memory. You are forced to re-explain requirements, paste data schemas, and manually review tickets.

While CLAUDE.md stores static hints, it is not a project management system or a live specification engine. As soon as multiple features are being built or business rules evolve, terminal memory gets fragmented.

ContextsBase turns Claude Code into a self-guiding execution engine:

• No Prompt Amnesia: Claude Code queries ContextsBase over MCP to fetch the current iteration and exact feature requirements.
• Zero Schema Guessing: Database models and API entities are delivered as rigid JSON schemas, preventing phantom fields.
• Automated Playwright Generation: Plain-text Given/When/Then test cases guide Claude Code to write and run actual end-to-end assertions.
• Drift Tracking: Once Claude Code finishes, it records an implementation fingerprint. If someone changes the business rules in ContextsBase later, the record flags Outdated automatically.

Anthropic’s Claude Code CLI provides a built-in command to register MCP servers directly into its configuration file:

Method 1: Using the Claude CLI (Recommended)

claude mcp add contextsbase \
  -- npx -y @contextsbase/mcp-server \
  --env CONTEXTSBASE_TOKEN="cb_tok_live_abc123..."

Method 2: Manual Config File (~/.claude.json)

If you manage configurations centrally, add ContextsBase directly to your ~/.claude.json or project-level configuration:

{
  "mcpServers": {
    "contextsbase": {
      "command": "npx",
      "args": ["-y", "@contextsbase/mcp-server"],
      "env": {
        "CONTEXTSBASE_TOKEN": "cb_tok_live_abc123..."
      }
    }
  }
}

Verify in the Terminal

Launch Claude Code in your terminal with claude and run /mcp. You should see contextsbase listed as connected with 5 active tools.

ContextsBase grants Claude Code precise, scoped access to the software requirements lifecycle:

Here is the standard workflow when pairing Claude Code with ContextsBase:

claude "Claim the next pending feature from ContextsBase, implement it according to its business rules, generate the Playwright test, and report results back."

Claude Code executes the pipeline autonomously:

1. 01
   Claim Step: Claude Code invokes next_iteration_item. The server locks the feature (e.g. F-2 Team Invites) atomically.
2. 02
   Code Implementation: Claude Code reads the schema definition and edits the codebase without guessing field names or edge case rules.
3. 03
   Playwright Verification: It translates the Given/When/Then test cases into Playwright scripts and runs them locally.
4. 04
   Sync Records: Claude Code calls record_test_automation and record_feature_implementation, moving the feature from In progress to Implemented.

Watch ContextsBase serve an agent, manage the queue, and record implementation state:

Many developers rely exclusively on CLAUDE.md. Here is how it compares to an MCP-native context server: