Summary
The current context7.instructions.md skill hardcodes Context7 (Upstash-hosted service) as the documentation retrieval MCP. This creates a dependency on a proprietary, externally hosted service whose backend index, parsing, ranking, and ingestion infrastructure are not open source. This issue proposes either replacing or making the skill tool-agnostic so it works equally well with open-source, self-hostable alternatives.
Problem
Context7's MCP server code is open source (upstash/context7), but its hosted platform (the documentation index, ingestion, parsing, ranking, and retrieval backend) is proprietary and operated by Upstash. This means:
- All documentation queries are sent to an external third-party service.
- There is no self-hosted option for the full backend.
- Enterprise/privacy-sensitive environments cannot use it safely for internal or confidential codebases.
- Teams with niche or private libraries get limited value from the pre-built public index.
Recommended open-source alternatives
| Tool |
Type |
Best for |
| GitMCP |
Open source, self-hostable |
Any public GitHub repository; converts repo docs/README/llms.txt into an MCP source |
| MCPDoc (langchain-ai/mcpdoc) |
Open source, self-hostable |
Documentation sites publishing llms.txt; developer-controlled and auditable |
| ContextMCP |
Open source, self-hostable |
Enterprise/internal knowledge bases with custom ingestion, chunking, and indexing |
| open-context7 |
Open source, self-hostable |
Privacy-first Context7 pattern without sending data to external servers |
Proposed change
Option A — Make the skill tool-agnostic: Rewrite context7.instructions.md to describe the intent (retrieve authoritative, version-specific documentation before writing code) and the tool workflow pattern, without naming a specific MCP service. Agents can be configured with Context7, GitMCP, MCPDoc, or any compliant documentation MCP depending on their environment.
Option B — Offer parallel skills: Keep context7.instructions.md as-is for those using the hosted service, and add:
gitmcp.instructions.md for GitMCP-based workflows
mcpdoc.instructions.md for MCPDoc-based workflows
Token efficiency note
Context7's newer architecture reportedly reduced retrieved documentation from ~9.7k to ~3.3k tokens average (65% reduction). Any replacement skill should preserve the efficiency guidance:
- Narrow task-specific queries only
- Retrieve minimal surrounding context
- Limit
resolve-library-id and query-docs equivalent calls per turn
- Skip external retrieval for local refactors and language fundamentals
References
Summary
The current
context7.instructions.mdskill hardcodes Context7 (Upstash-hosted service) as the documentation retrieval MCP. This creates a dependency on a proprietary, externally hosted service whose backend index, parsing, ranking, and ingestion infrastructure are not open source. This issue proposes either replacing or making the skill tool-agnostic so it works equally well with open-source, self-hostable alternatives.Problem
Context7's MCP server code is open source (upstash/context7), but its hosted platform (the documentation index, ingestion, parsing, ranking, and retrieval backend) is proprietary and operated by Upstash. This means:
Recommended open-source alternatives
llms.txtinto an MCP sourcellms.txt; developer-controlled and auditableProposed change
Option A — Make the skill tool-agnostic: Rewrite
context7.instructions.mdto describe the intent (retrieve authoritative, version-specific documentation before writing code) and the tool workflow pattern, without naming a specific MCP service. Agents can be configured with Context7, GitMCP, MCPDoc, or any compliant documentation MCP depending on their environment.Option B — Offer parallel skills: Keep
context7.instructions.mdas-is for those using the hosted service, and add:gitmcp.instructions.mdfor GitMCP-based workflowsmcpdoc.instructions.mdfor MCPDoc-based workflowsToken efficiency note
Context7's newer architecture reportedly reduced retrieved documentation from ~9.7k to ~3.3k tokens average (65% reduction). Any replacement skill should preserve the efficiency guidance:
resolve-library-idandquery-docsequivalent calls per turnReferences