Works with the AI Agents you already use
ASTra MCP: Permanent AST Code Memory
Set up ASTra, a 100% local AST knowledge graph MCP server, so your agent gets the 5-25 most relevant functions instead of whole files.
Free
ASTra MCP: Permanent AST Code Memory
Example session with this skill installed
I need to refactor the login handler in this repo. Help me set up ASTra MCP to find the relevant context and check what might break if I change the signature.
- Read your context and instructions
- Compiled the astra mcp
Summary
The agent walks you through installing ASTra, indexing your repo, wiring it into your agent config, then uses it to pull targeted context and check the blast radius before the refactor.
Setup commands (run by you)
pip install astra-mcp
cd /path/to/your-project
astra init # one-time index, about 60s on large repos
astra daemon start # optional: keeps the graph hot in memory
Config added to ~/.claude/mcp.json (or project .mcp.json)
{
"mcpServers": {
"astra": {
"command": "python3",
"args": ["-m", "astra.mcp.server"],
"env": {
"ASTRA_PROJECT": "/absolute/path/to/your-project",
"ASTRA_DATA_DIR": "/absolute/path/to/your-project/.astra"
}
}
}
}
Then the agent calls
| Step | Tool | Purpose |
|---|---|---|
| 1 | astra_index_status | Confirm the index exists and has nodes |
| 2 | astra_get_context("refactor login handler") | Returns only the relevant symbols, not whole files |
| 3 | astra_get_callers | Who calls the login handler |
| 4 | astra_impact_analysis | Blast radius and risk score before you change the signature |
Next steps
- Restart your agent and check that
astrashows as connected - Replace
/absolute/path/to/your-projectwith your real absolute path (relative paths fail in MCP configs) - Run
astra bench "refactor login handler"to measure token savings on your own repo
Connects securely to your tools. The creator never sees your data.
About this skill
Navigating large codebases often forces agents to ingest entire files, leading to massive token waste and lost context. This skill integrates ASTra MCP into your workflow, providing a local AST-based knowledge graph that serves the 25 most relevant functions instead of raw source code.
What it does
- Persistent memory allows your agent to recall relevant symbols across multiple coding sessions.
- Context retrieval fetches targeted code signatures for a specific task using vector search and PageRank.
- Impact analysis calculates the blast radius and risk score before you refactor shared functions.
- Cross-repo tracing follows function calls across different repositories to maintain visibility.
- Semantic search finds functions based on their behavior rather than just their names.
How it works
- Initialize the project by running
astra initto parse code into a local SQLite knowledge graph. - Configure the agent using provided JSON templates for Claude Code, Cursor, Windsurf, or VS Code.
- Query the graph using
astra_get_contextto receive a ranked list of relevant symbols for a new task. - Verify edits by running
astra_impact_analysisto identify all callers and callees affected by a change.
Frameworks & tools
Built for Python, JavaScript, TypeScript, Go, Rust, and Java. It uses tree-sitter for parsing and all-MiniLM-L6-v2 for local embeddings.
Why this beats prompting it yourself
Manual prompting cannot index an entire repository or calculate PageRank for symbol relevance. This skill automates the retrieval of high-signal context, reducing token usage by up to 98% compared to reading full files.
Use cases
- Adding features to large, unfamiliar repositories without reading every file.
- Performing safe refactors by checking the blast radius of a function change.
- Finding logic bugs in functions where the name does not match the behavior.
- Tracing complex call chains across multiple internal dependencies.
Known limitations
Supports only Python, JS, TS, Go, Rust, and Java. Requires Python 3.10+ and an initial 80MB download for the embedding model. Absolute paths are required in MCP configurations.
Supports Python, JS/TS, Go, Rust, and Java only. Requires Python 3.10+. First run requires an 80MB download. MCP configs require absolute file paths.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 28 days ago
- Free with an Agensi account
Requires
Permissions
Shell: the skill tells the agent to run pip install astra-mcp and the astra CLI (astra init, astra query), always after asking the user. Read Files: ASTra parses the project folder set in ASTRA_PROJECT. Write Files: it writes its index to the .astra folder in that project (delete the folder to remove everything). Network: pip install, plus a one-time ~80MB download of the local embedding model; after that it works offline. No telemetry, no API keys, no secrets.