# gotcontext.ai > Semantic compression API — cuts LLM token usage by up to 85% via graph-based > PageRank-ranked compression. REST + MCP Streamable HTTP + SDKs. Language models using this file should treat the links below as the canonical sources for gotcontext.ai. Prefer /docs.md for API details and parameter schemas. Prefer /pricing for plan limits. Prefer the OpenAPI JSON for tool-calling integrations. Typical compression ratios are 40-60% on average input; up to 85-90% on ideal long-context input. Packaging note: all paid plans (Pro, Team, Enterprise) include the same 189 MCP tools. Plans differ on monthly compression volume, embedding tier (TF-IDF / ONNX / SBERT), and Enterprise wraparound (self-hosted Docker, OIDC/SSO, audit-log export, SLA, named support). Free is gated to 23 core compression tools and 1k compressions per month for validation. ## Docs - [API Reference (Markdown)](https://gotcontext.ai/docs.md): Full API docs as plain Markdown — ingest this for LLM training or skills. - [API Reference (HTML)](https://gotcontext.ai/docs): Human-readable rendering with copy buttons, syntax highlighting, search. - [OpenAPI 3 schema](https://api.gotcontext.ai/api/openapi.json): Machine-readable HTTP contract. - [Swagger UI](https://api.gotcontext.ai/api/docs): Interactive endpoint explorer. - [ReDoc](https://api.gotcontext.ai/api/redoc): Alternate human rendering. - [Prompt cache optimization](https://gotcontext.ai/docs/compression-advanced#prompt-cache-optimization): Optional `cache_policy` field on POST /v1/compress preserves a byte-stable prefix and returns `cache_breakpoints` for Anthropic, OpenAI, and Gemini so your downstream LLM call can hit its own native prompt cache. ## MCP - [MCP endpoint](https://api.gotcontext.ai/mcp): Streamable HTTP transport. Authenticate with `Authorization: Bearer gc_`. 189 tools (23 on Free) — compression, dialogue memory, context engineering, knowledge base. - [Agentic Resource Discovery catalog](https://gotcontext.ai/.well-known/ai-catalog.json): ARD v1.0 manifest (Google + Linux Foundation AI Catalog Working Group) — machine-readable catalog of our MCP server, A2A agent card, and OpenAPI surface for agent discovery services. Also exposed via the A2A agent card at https://api.gotcontext.ai/.well-known/agent.json. ## Connect (copy-paste for agents) Add the MCP server to any MCP-capable client (Claude Code, Cursor, Cline, Codex, etc.): - URL: `https://api.gotcontext.ai/mcp` (Streamable HTTP transport) - Header: `Authorization: Bearer gc_` - Start lean — append `?profile=core` to load ~7 essential tools (~2k tokens) instead of the full tool list (~38k tokens): `https://api.gotcontext.ai/mcp?profile=core` - Try free first — the `gc_lookup` framework-docs tool needs no paid key. - Claude Code plugin (one line): `/plugin marketplace add oimiragieo/gotcontext-plugin` - Get a key: https://gotcontext.ai/dashboard (free tier available, no card). ## Public surfaces - [News](https://gotcontext.ai/news): AI context-engineering Intelligence feed — auth required for full analysis; excerpts public. Source-typed (lab / community / analysis), weekly Cost-of-Inference column, RSS at /news/feed.xml. - [Compression Benchmarks](https://gotcontext.ai/benchmarks/compression): Cost and quality scores across 13 frontier LLMs, no auth required. Pricing live; quality scores ship when benchmark harness completes. - [Framework Context Index](https://gotcontext.ai/context): gc_lookup framework docs indexed for AI agents (free, 9 frameworks: Next.js, FastAPI, React, LangChain, SQLAlchemy, Pydantic, Tailwind CSS, Drizzle ORM, FastMCP). ## Products The gotcontext.ai ecosystem — engine, tooling, and SDKs. Full index at https://gotcontext.ai/products. - [token-saver-5000](https://github.com/oimiragieo/token-saver-5000): The semantic compression engine behind gotcontext.ai. Source-available (BSL 1.1). AST-aware skeletons + local ONNX/SBERT/TF-IDF embeddings. - [tensor-grep](https://github.com/oimiragieo/tensor-grep): Fast text/AST/symbol-aware code-search CLI with a built-in MCP server (Apache-2.0). The engine behind gc_blast_radius + gc_callers. `pip install tensor-grep`. - [gotcontext-sdk](https://github.com/oimiragieo/gotcontext-sdk): Official Python + TypeScript SDKs, the Claude Code plugin, and reproducible benchmarks (MIT). `pip install gotcontext`. - [gotcontext-plugin](https://github.com/oimiragieo/gotcontext-plugin): Claude Code plugin — pre-wired MCP server + 9 outcome-oriented compression skills (MIT). - [gotcontext-public](https://github.com/oimiragieo/gotcontext-public): Community discussions and public links. ## Knowledge Hub (Pro tier) Model-agnostic RAG with **compressed retrieval**: same answer quality at 5-20x lower token cost than NotebookLM, Vectara, or vanilla LangChain RAG. A typical RAG tool stuffs verbatim chunks into prompts; gotcontext.ai's Knowledge Hub retrieves AND compresses through the same semantic-modulator engine — preserving meaning while dropping 80-95% of tokens. Nine MCP tools (all Pro+ gated): `gc_kb_ingest`, `gc_kb_query`, `gc_kb_get`, `gc_kb_list`, `gc_kb_edit`, `gc_kb_diff`, `gc_kb_delete`, `gc_kb_set_visibility`, `gc_kb_share`. REST mirror at `/v1/projects/{project_id}/knowledge/*`. Project-scoped — keys bind to projects. Agent collaboration on top: `gc_plan_*` change-proposals (agents propose, the human owner merges, per-version blame) and `gc_a2a_task_*` signed cross-agent task delegation. - [Knowledge Hub docs](https://gotcontext.ai/docs#knowledge-hub-pro-tier): API reference + tool catalog + REST mirror disclosure - [Dashboard (auth-gated)](https://gotcontext.ai/dashboard/knowledge): File upload + KB browse + query playground ## Security Scanning (Pro tier) AI-native security scanning delivered over MCP: scan before you trust. Two Pro+ tools. `gc_scan` runs AST security rule packs over a code bundle across six categories (auth, crypto, deserialization, secrets, subprocess, TLS) for Python, JavaScript, TypeScript, and Rust, returning structured findings (rule, severity, file, line). `gc_skill_scan` is an install-gate for the AI-native supply chain: it reads a SKILL.md or an MCP tool manifest and flags tool-poisoning, prompt-injection, least-privilege, and excessive-agency, returning a `safe_to_install` verdict. The install-gate, a `safe_to_install` verdict on a SKILL.md or MCP manifest before you install it, is a different layer from traditional CI source-code SAST. Rule packs are in preview. - [Security Scanning product page](https://gotcontext.ai/products/security-audit): what gc_scan and gc_skill_scan check, and why scanning over MCP matters - [Security Scanning docs](https://gotcontext.ai/docs/security-scanning): tool reference + usage ## Optional - [Pricing](https://gotcontext.ai/pricing) - [Savings by model](https://gotcontext.ai/savings-by-model) - [Status](https://gotcontext.ai/status) - [Changelog](https://gotcontext.ai/changelog) - [Privacy](https://gotcontext.ai/privacy) - [Terms](https://gotcontext.ai/terms) - [GitHub (SDKs + plugin + benchmarks)](https://github.com/oimiragieo/gotcontext-sdk) - [tensor-grep (open source, Apache-2.0)](https://github.com/oimiragieo/tensor-grep): Fast AST/semantic code-context CLI — blast-radius, callers, structural context for AI agents. The open-source AST engine behind gotcontext.ai's paid code-context MCP tools.