AI tools and tools for AI

Give your agents
memory that matters

CoffeeCode builds developer tools for AI-native workflows. Our first product lets agents look up anything that happened across every past session — decisions, changes, bugs, and why things were built the way they were.

☕ coffeectx MCP

Everything your agent ever decided, instantly accessible

coffeectx is a Model Context Protocol server that indexes all your agent session logs into a semantic knowledge graph. In a brand new session, your agent can query the full history of past decisions, bugs, architecture choices, and implementation changes — with no manual context-stuffing required.

Tested with Claude Code. Supports semantic search, regex, and graph-like queries. The knowledge graph is extensible: define new entity types and extraction skills to teach it what matters most to your project.

Explore on GitHub →
coffeectx — live queries
How did the MCP interface change over time?
Tell me about notable bugs and how they were solved.
How are log entries passed to the indexing agent, and which implementation was used before?
What architectural assumptions were made in the auth layer?
Find all decisions related to performance optimizations.

Queries that go deep

Three complementary search modes so your agent always finds exactly what it needs.

🔍
Semantic Search

Ask in plain language. coffeectx understands meaning, not just keywords — so "how was auth designed?" finds every relevant decision even if they used different words.

Regex Queries

Pin down exact patterns: function names, error codes, config keys. Fast, precise, and composable with semantic results.

🕸️
Graph Traversal

Navigate relationships between decisions, files, bugs, and features. Understand what caused what, and trace the full lineage of any choice.

🧩
Extensible Schema

Define your own entity types and extraction skills. Teach the graph what's important for your domain — deployment targets, experiment IDs, customer names, anything.

🤝
MCP Native

Plugs directly into the Model Context Protocol ecosystem. Any MCP-compatible agent can use coffeectx without custom integration code.

🔗
coffeeclaw Integration

Seamless first-party integration with coffeeclaw — giving agents persistent, queryable memory across every session out of the box.

From logs to knowledge in three steps

Index your sessions

Point coffeectx at your agent log directory. It reads every session, extracts entities and relationships, and builds an optimized knowledge graph.

coffeectx index ./logs
Connect via MCP

Add coffeectx as an MCP server in your agent config. One line of JSON. No API key, no cloud dependency, runs locally.

{ "mcpServers": { "coffeectx": { "command": "coffeectx", "args": ["serve"] } } }
Query from any session

Your agent can now ask natural-language questions about the entire project history from a fresh session — no manual context loading required.

ctx.query("Why was the MCP interface redesigned?")

Start brewing better AI memory

coffeectx is open source and free to use. Star the repo, try it with Claude Code, and help shape where it goes next.

View on GitHub