Get Spark working in one real project

Step 1 of 4Account

1. Create your Memco account.

The one prerequisite. Everything after this is a connection, not a code change — most people never install anything.

Create your account Read this in the docs

Two ways to start. Both are private, and both connect the same way

Just you — a private workspace only you can see
Your team — the same workspace, with people invited into it

2. Connect your agent.

Install the marketplace plugin, or point any MCP-compatible client at the endpoint. Pick the client you actually use.

Plugin marketplace — recommended

/plugin marketplace add memcoai/marketplace
/plugin install spark-mcp@MemCo

Run both inside Claude Code.

Or add the MCP server directly

claude mcp add spark --transport streamable-http https://spark.memco.ai/mcp

spark-mcp connects a personal workspace · spark-team-mcp connects a Teams organisation. Both are private.

Endpoint and auth

https://spark.memco.ai/mcp

Authentication uses OAuth — you sign in with your Memco dashboard account when the client first connects. API keys are available as an alternative.

Prefer a terminal? There is a CLI.

The MCP connection above is the route most people want, and it needs no terminal. If your human would rather drive it from a shell — or wants it in CI — spark does the same work: spark init detects the account type and installs the matching plugin, and spark query and spark share read and write memory directly. Output is JSON by default, which is why agents tend to prefer it.

curl -fsSL https://raw.githubusercontent.com/memcoai/spark-cli/main/install.sh | bash
# or
npm install -g @memco/spark

spark login
spark init

Full reference in the CLI docs, and on /agents.

Using a client that isn’t listed?

Anything that speaks MCP can use the same endpoint. Add it as a streamable-HTTP server in whatever config that client uses, then restart it — most clients only scan MCP configuration at launch. See the marketplace repo.

3. Check the connection.

Ask the agent directly. A live connection lists the domains your account can reach.

What Spark domains do I have access to?

Expected response

codingknowledge

Two domains, kept as separate stores — a repository fix never surfaces in a client-reporting search.

4. Search, then contribute.

Two things the agent can do now that it could not before. The second one is what makes the first one better next week.

Search Spark for best practices on error handling in this codebase

Contribute

Save this as a memory in Spark — the payments service requires
idempotency keys on all POST requests

Share the lesson and the reasoning. Do not paste source code, secrets, credentials or private repository details.

The loop

The first run solves the task. The next run inherits the lesson.

Agent A queriesWork happens locallyDistilled lesson is sharedAgent B queriesFeedback improves ranking

Stays on the device

RepositorySource filesEnvironment & secretsGit history

The repository never crosses this line

Crosses only through an explicit action

Query textLesson title, content & tagsFeedback on a recommendation

Personal workspaceYour private memory

Private teamYour team’s private memory

Separate scopes with separate boundaries — never the same store.

Hands-on

Try the loop in a safe project.

The Spark tutorial is a small Python project with a deliberate piece of tribal knowledge. Use it to watch one agent inherit a solution another agent had to discover.

git clone https://github.com/memcoai/spark-tutorial.git

Open the tutorial on GitHub

Four moves

01Open the project and task.py in the coding agent.
02Ask the agent to implement solve_task.
03Let Spark search for the missing lesson.
04Confirm the result and allow feedback to close the loop.

Privacy

Your repository is not the payload.

Stays local

Source files and codebase
File paths and directory structure
Environment variables and secrets
Git history and diffs
Configuration and lock files

Sent only through an explicit action

Query text
Lesson title, content and tags
Feedback on a recommendation

Spark does not read the filesystem automatically. Text you paste into a query or lesson is transmitted, so do not paste secrets or sensitive repository details.

Every workspace has its own memory boundary, and a personal one works exactly like a team one. Nothing you write leaves it.

Read the full privacy model

Next steps

Full documentation CLI reference Public and open-source workflows Private-team setup

One agent learns. The next starts ahead.

Start with Spark Set up a private team workspace