Your AI agents don’t sleep. Now neither does their memory.

Step away from your desk and productivity drops to zero. AgentiBridge makes your Claude Code sessions persistent, searchable, and remotely controllable — from any MCP client.

Get Started View on GitHub

The Problem

Your Claude Code sessions disappear when the terminal closes. Your IDE history is locked to one machine. When you leave your desk, your agent productivity drops to zero.

AgentiBridge fixes this.

It indexes every Claude Code transcript automatically, makes them searchable with AI-powered semantic search, and lets you dispatch and monitor tasks from any device — your phone, another laptop, or claude.ai.

AgentiBridge vs Anthropic Remote Control

Anthropic’s Remote Control (Feb 2026) continues one live session from your phone. AgentiBridge is the other 90% of remote agent control:

	Remote Control	AgentiBridge
What it does	Continue one live session from phone/browser	Index all past sessions, search, dispatch new ones
Session scope	1 active session	All sessions across all projects
Search	None	Keyword + semantic search
Memory/Plans	None	Catalog memory files, plans, history
Dispatch	None	Fire-and-forget background task dispatch
Context restore	Only the live session	Restore any past session’s context
Multi-client	claude.ai + Claude app only	Claude Code, claude.ai, ChatGPT, Grok, any MCP client
Runs where	Local terminal (must stay open)	MCP server (Docker/stdio), works offline
Auth	Pro/Max subscription	Self-hosted, no subscription needed

Bottom line: Remote Control is a “remote desktop for one session.” AgentiBridge is a knowledge base and orchestration layer across all sessions. They’re complementary, not competitive.

See It In Action — `agentibridge search`

One command from any shell. No interactive Claude Code session. No context switch.

agentibridge search "what was I doing in the email-template session around 20:00?"

Spawns a headless claude -p --model opus under the hood, wraps your query in a recon prompt, streams every tool call live to your terminal, and hands you back a human-readable summary with a claude --resume <session_id> footer so you can jump in and continue. Also available as the agent_search MCP tool for in-client use.

claude.ai using AgentiBridge list_sessions tool

AgentiBridge works from Claude Code CLI, claude.ai, ChatGPT, and any MCP client. See more screenshots →

Features

Agentic Search

agentibridge search "<q>" spawns a headless Opus one-shot that reasons over your sessions, history, memory, and git — streams live progress, returns a human-readable summary.

Dispatch + Handoff

Fire-and-forget task dispatch with session restore. Seed a fresh conversation in any project with structured context. Resume work where you left off, from any device.

Security First

OAuth 2.1 with PKCE, API key auth, Cloudflare Tunnel. Fully self-hosted — your data never leaves your infrastructure.

A2A + Multi-Client

Built-in Agent-to-Agent registry — agents register, heartbeat, discover peers by capability. Works with Claude Code CLI, claude.ai, ChatGPT, Grok, and any MCP client.

Quick Start

pip install agentibridge

AgentiBridge supports two connection modes. Pick one or use both.

Local (stdio) — Zero config

Runs as a subprocess alongside Claude Code. No server, no auth. Add to .mcp.json:

{
  "mcpServers": {
    "agentibridge": {
      "command": "python",
      "args": ["-m", "agentibridge"]
    }
  }
}

Remote (HTTP + API key) — Access from anywhere

Runs as a persistent server. Access your sessions from your phone, another laptop, or claude.ai:

{
  "mcpServers": {
    "agentibridge": {
      "type": "http",
      "url": "https://bridge.yourdomain.com/mcp",
      "headers": {
        "X-API-Key": "sk-ab-your-api-key-here"
      }
    }
  }
}

Both at once

Local for speed, remote for mobility — run them side by side:

{
  "mcpServers": {
    "agentibridge": {
      "command": "python",
      "args": ["-m", "agentibridge"]
    },
    "agentibridge-remote": {
      "type": "http",
      "url": "https://bridge.yourdomain.com/mcp",
      "headers": {
        "X-API-Key": "sk-ab-your-api-key-here"
      }
    }
  }
}

That’s it. Your Claude Code sessions are now searchable from any MCP-compatible client.

Zero dependencies to start — filesystem-only storage out of the box. Add Redis for caching and Postgres for semantic search when you need them.

MCP Tools

Foundation

Tool	What it does
`list_sessions`	List sessions across all projects
`get_session`	Full session metadata + transcript
`get_session_segment`	Paginated/time-range transcript retrieval
`get_session_actions`	Extract tool calls with counts
`search_sessions`	Keyword search across transcripts
`collect_now`	Trigger immediate collection

AI-Powered

Tool	What it does
`search_semantic`	Semantic search using embeddings
`generate_summary`	Auto-generate session summary via LLM
`agent_search`	Agentic recon — spawn headless Opus one-shot, return structured matches

Dispatch + Handoff

Tool	What it does
`restore_session`	Load session context for continuation
`dispatch_task`	Fire-and-forget background job dispatch
`get_dispatch_job`	Poll a background job for status and output
`plan_task` / `execute_plan` / `get_dispatch_plan` / `list_dispatch_plans`	Plan-first workflows for longer tasks
`list_handoff_projects` / `handoff`	Seed a conversation in another project with structured context

Knowledge Catalog

Tool	What it does
`list_memory_files`	List memory files across projects
`get_memory_file`	Read a specific memory file
`list_plans`	List plans sorted by recency
`get_plan`	Read a plan by codename
`search_history`	Search the global prompt history

Agent-to-Agent Registry

Tool	What it does
`register_agent` / `heartbeat_agent` / `deregister_agent`	A2A lifecycle
`list_agents` / `get_agent` / `find_agents`	Discover peers by type, status, or capability

Architecture

flowchart LR
    E([Any AI Client]) -->|query| D
    D -->|read| C[(SessionStore)]
    C -->|indexed by| B[Collector]
    B -->|watches| A([Claude Code sessions])

    classDef sessions fill:#6366f1,stroke:#4338ca,color:#fff
    classDef collector fill:#f59e0b,stroke:#d97706,color:#fff
    classDef store fill:#10b981,stroke:#059669,color:#fff
    classDef tools fill:#8b5cf6,stroke:#7c3aed,color:#fff
    classDef client fill:#06b6d4,stroke:#0284c7,color:#fff

    class A sessions
    class B collector
    class C store
    class D tools
    class E client

Connect to Claude.ai

Claude.ai requires OAuth 2.1 to connect to remote MCP servers. AgentiBridge includes a built-in OAuth 2.1 authorization server — enable it with one env var.

1. Add to your .env:

OAUTH_ISSUER_URL=https://bridge.yourdomain.com

2. Expose over HTTPS (Cloudflare Tunnel or reverse proxy):

agentibridge tunnel setup

3. Add to claude.ai at claude.ai/settings/connectors:

https://bridge.yourdomain.com/mcp

Claude.ai automatically discovers OAuth metadata, registers as a client, and completes the PKCE flow. No manual JSON config needed.

For production, set OAUTH_CLIENT_ID and OAUTH_CLIENT_SECRET to lock down registration to a single pre-configured client. API key auth (X-API-Key) continues to work alongside OAuth.

Deployment Options

AgentiBridge ships as a pip package only — no Docker image for the app. The only optional Docker footprint is Redis + Postgres sidecars managed by agentibridge install.

	Minimal	Standard	Production
Install	`pip install agentibridge`	`pip install agentibridge && agentibridge install`	`agentibridge install` + `agentibridge tunnel setup`
App	Native Python	Native Python (systemd user unit)	Native Python (systemd user unit)
Storage	Filesystem only	Redis + filesystem	Redis + Postgres (pgvector)
Search	Keyword only	Keyword only	Keyword + semantic + `agent_search`
Access	Local only	Local network	Internet (HTTPS via Cloudflare Tunnel)
Auth	None	API key	OAuth 2.1 + API key
A2A	Filesystem fallback	Redis-backed	Redis-backed, discoverable over HTTPS

FAQ

Isn't this just session history?

History is the data layer. The product is remote fleet control — dispatch tasks from your phone, search sessions from any MCP client, monitor jobs from claude.ai. You go from 0% productivity away from your desk to controlling your agents from anywhere.

VS Code / Cursor already has conversation history.

IDE conversation history is excellent for local replay within that IDE. AgentiBridge serves CLI-first developers and adds capabilities no IDE provides: remote multi-client access, background dispatch from any device, and semantic search across your full session history.

Won't Anthropic build this natively?

AgentiBridge is self-hosted, vendor-neutral infrastructure. Native features optimize for one vendor’s client. AgentiBridge works with Claude Code, claude.ai, ChatGPT, Grok, and any MCP client. Your data stays on your machine. MIT licensed — no lock-in.

Do I need Redis and Postgres?

No. pip install agentibridge && python -m agentibridge works with zero dependencies — filesystem-only storage out of the box. agentibridge install adds Redis + Postgres sidecars via Docker Compose for caching and semantic search. You can also point at an existing Redis/Postgres by setting REDIS_URL / POSTGRES_URL in ~/.agentibridge/agentibridge.env.

Is there a Docker image for AgentiBridge?

No. As of v0.5.0 AgentiBridge is a pip package only — the app runs natively via systemd. We dropped the app container image to shrink the surface area: one artifact (PyPI), one env file (~/.agentibridge/agentibridge.env), one unit to restart. Only the optional Redis/Postgres sidecars still use Docker.

Is my data sent anywhere?

No. No telemetry, no SaaS dependencies. Cloudflare Tunnel is opt-in, and even then only MCP tool responses traverse the tunnel — your transcripts stay local.

How is this different from Anthropic's Remote Control?

Remote Control continues one active Claude Code session from your phone — great for live interaction. AgentiBridge indexes all past sessions, adds semantic search, background dispatch, context restore, and multi-client support. They solve different problems and work great together.

Which clients are supported?

Claude Code CLI, claude.ai, ChatGPT, Grok, and any MCP-compatible client. Run agentibridge connect for ready-to-paste configs.

Code Quality

Continuous static analysis via SonarQube ensures code quality, security, and maintainability.

Ready to make your AI agents persistent?

pip install agentibridge

Get Started Documentation View on GitHub

Your AI agents don’t sleep. Now neither does their memory.

The Problem

AgentiBridge vs Anthropic Remote Control

See It In Action — agentibridge search

Features

Agentic Search

Dispatch + Handoff

Security First

A2A + Multi-Client

Quick Start

Local (stdio) — Zero config

Remote (HTTP + API key) — Access from anywhere

Both at once

MCP Tools

Foundation

AI-Powered

Dispatch + Handoff

Knowledge Catalog

Agent-to-Agent Registry

Architecture

Connect to Claude.ai

Deployment Options

FAQ

Code Quality

See It In Action — `agentibridge search`