- Category
- Developer Tools
- Rank
- No. 3622Tools index
Previous survey · No. 3478 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Coding
- Interfaces
- MCP
- GitHub
- 7 stars
- Latest release
- v4.1.0
- Date
About
An MCP server that parses code into a Tree-sitter AST in memory and prunes that context before passing it to a coding agent such as Cursor or Claude, cutting token usage by roughly 72%. It functions as a firewall between a codebase and an agent's context window, aiming to reduce cost and irrelevant context without losing structurally important code.
At a glance
Ctxfw provides an MCP path that can return code interfaces in place of full implementations. It is worth evaluating if agents spend too much context on peripheral modules and you can measure which details they lose.
Good fit
Teams testing context reduction in a supported repository with disposable inputs.
Consider something else when
Use a full-source view when a task depends on implementation details across multiple modules.
Start here
Choose one small repository and a question whose answer depends on a nearby module. Compare the pruned result and the full source for both token use and answer accuracy.
Cost and requirements
Python package for MCP integrations. The project publishes an Apache-2.0 core and advertises separate commercial components.
Limits to know
Pruning changes what the model sees. The project’s security and token-reduction claims do not establish coverage for your inputs. Keep application permissions and authorization checks in place.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Cuts token usage for coding agents by pruning context at the AST level instead of naive truncation, which could materially lower cost in long agent sessions if the claimed reduction holds up.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.