A CLI tool that dynamically prunes and caches MCP tool schemas per request to cut agent context bloat
Every MCP connection injects its full tool schema into the context window before any agent reasoning happens. Connecting 5 servers burns 50k+ tokens on definitions alone, 40-60% of a typical context budget gone before the first tool call. Existing fixes are partial: Atlassian mcp-compressor does static compression but cannot do dynamic per-request selection; Claude Code Tool Search cuts only 46.9%; Context Mode cannot intercept third-party MCP tool responses. This CLI proxy sits between the agent and all connected MCP servers, dynamically selecting only the schema slice relevant to the current request using lightweight semantic matching, caching permission profiles so read-only tools auto-approve, and compressing the remaining schema at request time. The result is a single local proxy that cuts context overhead by 80-95% and eliminates approval friction for pre-cleared tool categories without requiring changes to individual MCP servers or the host client.
Demand Breakdown
Social Proof 5 sources
Gap Assessment
5 tools exist (Context Mode, mcp-compressor (Atlassian Labs), mcp2cli, Claude Code Tool Search, MCPlexor) but gaps remain: Cannot intercept third-party MCP tool responses via JSON-RPC. Only built-in tools benefit. No permission profile layer.; No dynamic per-request selection based on task context. No permission profile layer. Open source with no monetization..
Features2 agent-ready prompts
Competitive LandscapeFREE
| Product | Does | Missing |
|---|---|---|
| Context Mode | Compresses Claude Code context usage by up to 98% for built-in tools like Bash and WebFetch. | Cannot intercept third-party MCP tool responses via JSON-RPC. Only built-in tools benefit. No permission profile layer. |
| mcp-compressor (Atlassian Labs) | Static schema compression achieving 70-97% token reduction on MCP tool definitions. | No dynamic per-request selection based on task context. No permission profile layer. Open source with no monetization. |
| mcp2cli | Full CLI bridge converting MCP to shell commands, achieving 96-99% fewer tokens than native MCP. | Breaks existing MCP-native workflows and requires rewriting tool integrations. Not a transparent proxy. |
| Claude Code Tool Search | Native Anthropic feature that defers tool schemas and loads them on demand in Claude Code. | Only 46.9% reduction. Claude Code-only, not cross-client. No permission caching. Cannot be used standalone. |
| MCPlexor | Static MCP router that appeared on HN with demand signal. | No dynamic selection or permission profiles documented. No distinct architectural differentiation from static compression. |
Notable VoicesFREE
"Measured 40-50% context window consumption from tool catalogs alone in MCP setups"
"315 KB of MCP output reduced to 5.4 KB in Claude Code; shipped and hit HN front page twice (Feb and Mar 2026)"
"96-99% fewer tokens than native MCP via CLI bridge approach"
Leads148BUILDER
Sign in to unlock full access.