Skip to main content

Sequential Thinking MCP

zerodb-sequential-thinking-mcp gives AI agents persistent, structured reasoning that survives across sessions, machines, and agent restarts. Built on ZeroDB plan artifacts.

npx zerodb-sequential-thinking-mcp

No account required. The first run automatically provisions a ZeroDB account. The second run is silent.

Why This Exists​

The upstream sequential-thinking MCP is stateless — every session starts fresh. This server adds:

  • Persistence — reasoning chains written to ZeroDB plan artifacts, survives restarts
  • Cross-session resume — Agent B can pick up where Agent A left off using an artifact_id
  • Self-provisioning — no account signup required to install and run
  • Multi-agent sharing — multiple agents share the same reasoning namespace via ZERODB_PROJECT_ID

Tools​

sequential_think​

Add a reasoning step to a chain.

Parameters:

NameTypeRequiredDescription
thoughtstringyesThe reasoning step content
chain_idstringnoChain identifier (auto-generated if omitted)
step_numbernumbernoExplicit step number
total_stepsnumbernoEstimated total steps
next_thought_neededbooleannoWhether more steps are expected

Returns: { chain_id, step_number, thought }

Example:

{
  "thought": "The root cause is connection pool exhaustion — PgBouncer is capped at 20 connections but we have 4 instances each holding 10.",
  "chain_id": "debug-pool-issue",
  "step_number": 2,
  "total_steps": 5,
  "next_thought_needed": true
}

sequential_conclude​

Finalize a reasoning chain and persist it as a ZeroDB plan artifact. Returns an artifact_id that can be used to resume the chain in a future session.

Parameters:

NameTypeRequiredDescription
chain_idstringyesChain to conclude
conclusionstringyesFinal conclusion or decision

Returns:

{
  "artifact_id": "cd6e95d9-0260-4f04-b817-ee461cf367f8",
  "step_count": 5,
  "resume_hint": "Use sequential_resume with artifact_id=\"cd6e95d9...\" to continue in a future session"
}

The artifact_id is stable — save it to resume this chain from any session or agent.

sequential_resume​

Load a prior reasoning chain by artifact ID. Use at the start of a session to continue where a previous agent left off.

Parameters:

NameTypeRequiredDescription
artifact_idstringyesArtifact ID from a previous sequential_conclude

Returns: Full prior reasoning chain content and step count.

Self-Provisioning​

The first time the server boots with no credentials configured, it calls:

POST https://api.ainative.studio/api/v1/public/instant-db

This creates a temporary ZeroDB account (~800ms) and writes credentials atomically to ~/.zerodb/credentials.json. Subsequent boots load from the file silently.

Credential Lifecycle​

StateBehavior
TemporaryAuto-provisioned, expires in 7 days
ClaimedUser claimed via link, permanent
ExpiredRe-provisioned automatically on next boot

To claim your account and make it permanent, set ZERODB_EMAIL:

{
  "env": {
    "ZERODB_EMAIL": "you@example.com"
  }
}

The server sends a claim link to your email on next boot.

Bring Your Own Credentials​

If you already have a ZeroDB account, skip provisioning entirely:

{
  "env": {
    "ZERODB_API_KEY": "zdb_live_your-key",
    "ZERODB_PROJECT_ID": "your-project-id"
  }
}

Configuration​

Claude Code​

{
  "mcpServers": {
    "sequential-thinking": {
      "command": "npx",
      "args": ["-y", "zerodb-sequential-thinking-mcp"]
    }
  }
}

With existing ZeroDB account​

{
  "mcpServers": {
    "sequential-thinking": {
      "command": "npx",
      "args": ["-y", "zerodb-sequential-thinking-mcp"],
      "env": {
        "ZERODB_API_KEY": "zdb_live_your-key",
        "ZERODB_PROJECT_ID": "your-project-id",
        "ZERODB_BASE_URL": "https://api.ainative.studio"
      }
    }
  }
}

Multi-agent swarm (shared reasoning)​

{
  "mcpServers": {
    "agent-alpha": {
      "command": "npx",
      "args": ["-y", "zerodb-sequential-thinking-mcp"],
      "env": {
        "ZERODB_CONFIG_DIR": "/shared/zerodb",
        "ZERODB_PROJECT_ID": "shared-project-id"
      }
    },
    "agent-beta": {
      "command": "npx",
      "args": ["-y", "zerodb-sequential-thinking-mcp"],
      "env": {
        "ZERODB_CONFIG_DIR": "/shared/zerodb",
        "ZERODB_PROJECT_ID": "shared-project-id"
      }
    }
  }
}

Both agents share the same credentials file and project namespace. Agent A's concluded chains are immediately resumable by Agent B.

Cross-Session Example​

Session 1 — Agent A investigates:

sequential_think("The API is returning 503s. Checking Kong logs...")
sequential_think("Kong http-log plugin is retrying on 404 — creating a retry storm.")
sequential_think("Disabling the plugin stops the storm. Root cause confirmed.")
sequential_conclude(chain_id="kong-debug", conclusion="Disable http-log plugin, redeploy Kong.")
→ artifact_id: "cd6e95d9-0260-4f04-b817-ee461cf367f8"

Session 2 — Agent B implements the fix:

sequential_resume(artifact_id="cd6e95d9-0260-4f04-b817-ee461cf367f8")
→ Returns Agent A's full reasoning chain
sequential_think("Updating kong.yml to remove http-log plugin...")
sequential_conclude(chain_id="kong-fix", conclusion="Deployed. 503s resolved.")