AI MEMORY FOR AGENTS · GOOGLE ADK
Give your next ADK run a stronger starting point.
We bring lasting project knowledge into your Google ADK workflow. Find the decision your next step needs, retrieve its original record, then save a clear handoff for the work ahead. Our example runs through ADK’s own agent, runner and MCP tools.
Start with our local, fixed-workflow example. No Gemini key, Google Cloud account or CMEM Pro account required for this example.

KNOWLEDGE THAT LASTS
Carry the decision into the next step.
Start with the decision already made.
We search your selected project and retrieve the original observation before the agent saves its next handoff.
Keep the handoff connected to its run.
We retain the logical run ID and actual ADK session ID with the saved note, giving the next step a record to return to.
Repeat a handoff without duplicating it.
The same project, run ID and note reuse the saved record. Your application decides when a new handoff is needed.
ILLUSTRATIVE HANDOFF · ACTUAL MCP WORKFLOW
Recall the decision. Save what comes next.
“The scheduler writes dates in UTC.”
Start with a known project observation. After the agent retrieves its original record, save a handoff such as “Next run: keep the scheduler export in UTC.”
- 01
Choose the known decision
Select an observation from your project and a query that finds it in the first 20 search results.
- 02
Read the original
The agent retrieves the selected note’s timeline and details through the native MCP tools.
- 03
Save the next step
Only after recall succeeds, the agent saves the handoff text you supplied for this project.
- 04
Keep the confirmed record
Receive the saved observation ID and ADK session ID. The note remains after the in-memory session ends.
AI memory in production
70 days of work. One searchable memory.
Alessandro used our open-source engine across three machines for 70 days. His customized deployment captured the decisions, fixes and discoveries behind the work—and published the data.
- Observations captured
- 22,718
- Bug fixes recorded
- 1,262
- Decisions recorded
- 757
See how he made that knowledge searchable. Then give your own agents a memory to build on.

START WITH ONE PROJECT
Recall one decision. Save the next step.
- Start your configured local worker. Follow the installation guide. From a Claude-Mem source checkout, inspect
npm run worker:statusand usenpm run worker:startif needed. - Install the pinned Python environment. Use Python 3.11 or later and Node.js. Run below from the source checkout; dependencies pin Google ADK 2.14.0 and MCP 2.3.0. Set the absolute installed or shipped MCP script path.
- Choose a real observation in your project. Replace
123with its ID. Changeshop-apiandquokkaharborto your project and a query that finds that ID in the first 20 results. The full guide includes an optional harmless seed-note example.
python3 -m venv .venv-adk-memory
. .venv-adk-memory/bin/activate
python -m pip install -r examples/google-adk/requirements.txt
export CLAUDE_MEM_MCP_SERVER="/absolute/path/to/claude-mem/plugin/scripts/mcp-server.cjs"
python examples/google-adk/memory_workflow.py \
--project shop-api \
--query quokkaharbor \
--observation-id 123 \
--run-id scheduler-handoff-1 \
--note "Next run: keep the scheduler export in UTC."Match custom CLAUDE_MEM_DATA_DIR and CLAUDE_MEM_WORKER_PORT values to your worker. The MCP child uses local worker mode with autostart disabled; this example adds no hooks or new MCP configuration.
Check success. Output includes the recalled observation, saved_id and adk_session_id. Confirm the original decision was recalled and the intended handoff saved. The same project, run ID and note return the same saved ID on a deliberate repeat.
Keep the four-tool filter, explicit project scope and error checks when adapting the agent. Recalled text cannot change scope or authorize another action. This example sends no note to a model; existing worker sync and export settings still apply.
Before you connect
Does this need Gemini or Google Cloud?
Not this custom fixed workflow. It uses the real ADK runner and MCP toolset without a model call. Model-driven applications have their own provider requirements. No CMEM Pro account is needed for this local example.
Which note can I recall?
Select an observation belonging to your chosen project. Its ID must appear in the first 20 search results; narrow the query if necessary. The agent reads its timeline and details before saving the handoff.
Does it capture every conversation?
No. It retrieves selected memory and saves caller-supplied handoff text. No transcript hooks or automatic context injection are added.
What survives the run?
The explicit note and provenance persist in Claude-Mem. The example’s ADK session is in memory and discarded; its ID is recorded as metadata. The local example user identity is not authentication to a remote account.
How do repeated runs work?
The fingerprint uses the integration, project, run ID and note. Identical inputs reuse the saved ID; a changed note is a new save. Use a new logical run ID or changed note when intentionally saving a new handoff. Query or selected source ID alone does not change that fingerprint.
What if a request fails?
The CLI fails with a content-free reason. Requests default to 10 seconds and workflow execution to 80 seconds. Failed recall stops the handoff. No automatic write retry runs; inspect saved memory before repeating an uncertain write.
Which versions and platforms are verified?
The published checks use Python 3.12.14, Google ADK 2.14.0, MCP Python SDK 2.3.0 and Claude-Mem 15.16.1 on Linux. macOS, Windows, ADK Web, cloud deployments, remote authentication and model-selected tool calls were not verified by this fixture.
How do project boundaries and disconnect work?
Reads carry the selected project and a one-item projects list. These filters prevent worktree expansion; they are not multi-user account authorization. Stop running or remove the example to disconnect. The toolset and runner close after the workflow; removing its virtual environment does not erase saved notes. Existing worker privacy, sync and export settings still apply.
Build on a working example.
Source reviewed November 20, 2026. The published nine-check fixture exercises the actual ADK runner, custom agent, MCP toolset and local worker. It checks known-note recall, handoff save, deliberate repeats, project and worktree boundaries, provenance, and expected failures. This is not a model-quality benchmark.
Keep the decision. Move the work forward.
Give your ADK workflow a source it can verify and a handoff the next run can use.