CCClaude Code project memory

Persistent project memory for Claude Code.

Give Claude Code the repo knowledge, decisions, lessons, and handoffs your last session already earned. ContextStream combines durable project memory with code search and dependency context so the next session can start from evidence instead of a blank slate.

Install or upgrade the ContextStream MCP server

macOS / Linux$curl -fsSL https://contextstream.io/scripts/mcp.sh | bash
Windows PowerShell$irm https://contextstream.io/scripts/mcp.ps1 | iex

Starter: free with 10,000 credits/mo · Pro: 25,000 credits/mo

01What Claude Code can carry forward

Start the next session warm.

ContextStream gives Claude Code a structured place to retrieve the facts that do not belong in a one-off chat transcript.

01

Codebase evidence

Search indexed code and inspect dependency, call, dataflow, and type relationships before changing a high-blast-radius module.

02

Decisions and lessons

Retrieve why the team chose an approach, which hypotheses were eliminated, and which corrections should not be repeated.

03

Plans and handoffs

Keep implementation steps, verification state, and actionable next work available when a session or coding tool changes.

A REAL FIRST USE

Ask what the project already knows.

› Before editing auth retries, find the decision behind the current policy and show which code depends on it.

Claude Code can retrieve the relevant project memory and source evidence, then explain what it found before it edits.

PLAN CONTEXT

Full featured from Free. More credits when you need them.

Starter includes 10,000 Context Credits each month, unlimited workspaces, up to 15 projects, and persistent project memory. Pro starts at $19/month with 25,000 monthly credits and unlimited projects. Elite is $49/month with 100,000 monthly credits. Every plan includes agents, orchestration, coordination, decision learning and full code graphs.

02Published evaluation

Measure retrieval, not marketing adjectives.

Read the methods, datasets, judging criteria, and known losses behind these ContextStream results.

Review benchmarks →
90.0%
LONGMEMEVAL-S ACCURACY
99.38%
CODE SEARCH RECALL@10

Give Claude Code the context your project already earned.

Install the MCP server, connect one real project, and test the next session against decisions and code you already know.