Context7 MCP

v2026.6

Upstash

MCPdocumentationcode-contextretrievalupstash
79
Strong
About This MCP

Upstash's documentation-retrieval MCP server. Two tools (resolve-library-id, get-library-docs) inject up-to-date, version-specific library documentation into the agent's context to prevent hallucinated APIs. The most-starred MCP server repo (57.1k), but with a notable security history: the ContextCrush content-injection vulnerability (disclosed Feb 2026, patched within days).

Last Evaluated: June 10, 2026
Official Website

Trust Vector Analysis

Dimension Breakdown

🚀Performance & Reliability
+
documentation accuracy

Spot-check of returned documentation against official library docs across popular frameworks

Evidence
Context7 READMEDocs are parsed from official library sources and served version-specifically, directly addressing hallucinated or outdated APIs in LLM code generation
highVerified: 2026-06-10
retrieval relevance

Relevance assessment of topic-filtered retrievals across common and long-tail queries

Evidence
Context7 get-library-docs toolTopic-scoped retrieval with configurable token budget returns focused snippets; relevance degrades for niche topics in very large libraries
mediumVerified: 2026-06-10
api reliability

Availability monitoring of the hosted API endpoint over the evaluation period

Evidence
Context7 serviceBackend hosted on Upstash infrastructure; remote Streamable HTTP endpoint and local stdio server both depend on the hosted API, which has shown solid availability
mediumVerified: 2026-06-10
library coverage

Coverage sampling across mainstream and long-tail open-source libraries

Evidence
Context7 library indexTens of thousands of indexed libraries with community submission of new ones; resolve-library-id reliably maps natural-language names to indexed library IDs
highVerified: 2026-06-10
error recovery

Error-path testing with unknown libraries, rate limits, and offline backend

Evidence
Context7 MCP implementationUnresolved library names return candidate matches rather than hard failures; rate-limit and backend errors surface as readable messages the agent can react to
mediumVerified: 2026-06-10
🛡️Security
+
content injection resistance

Review of the ContextCrush vulnerability, its patch, and the residual risk of doc-content prompt injection

Evidence
Noma Labs — ContextCrush disclosureContextCrush: the Custom Rules / 'AI Instructions' feature let anyone publishing a library inject unsanitized instructions into consuming agents' context; researchers demonstrated .env credential theft. Disclosed 2026-02-18, patched 2026-02-23
Context7 repositoryPost-patch, publisher-supplied instruction content is sanitized/restricted, but retrieved documentation remains third-party content flowing into the agent context by design
highVerified: 2026-06-10
supply chain trust

Analysis of the open library-submission pipeline as an attack surface for agent contexts

Evidence
Noma Labs — ContextCrush disclosureAnyone can publish or update a library in Context7's index, making indexed docs an open supply chain into agent contexts; ContextCrush proved this channel was exploitable at scale
highVerified: 2026-06-10
vulnerability response

Assessment of disclosure-to-patch timeline and vendor cooperation

Evidence
Noma Labs — ContextCrush disclosure timelineDisclosed to Upstash 2026-02-18; patched 2026-02-23 (5 days); coordinated public disclosure 2026-03-05 — fast remediation, though the vulnerable feature had shipped without sanitization review
highVerified: 2026-06-10
credential exposure risk

Analysis of indirect credential-theft pathways via injected instructions in a multi-tool agent

Evidence
Noma Labs — ContextCrush demonstrationThe demonstrated exploit exfiltrated .env credentials by instructing the agent to read and transmit secrets — Context7 itself holds no credentials beyond an optional API key, but its content channel could weaponize other tools the agent holds
highVerified: 2026-06-10
authentication security

Review of API-key handling for the hosted endpoint and local server

Evidence
Context7 documentationWorks without authentication for basic use; optional API key raises rate limits and is passed via header/env var. Minimal credential surface, but the remote endpoint means key handling depends on client configuration
mediumVerified: 2026-06-10
🔒Privacy & Compliance
+
query data exposure

Data flow analysis of outbound query content to the hosted backend

Evidence
Context7 architectureLibrary names and topic queries are sent to Upstash's hosted API; queries can reveal what technologies a team is using but contain no source code
mediumVerified: 2026-06-10
sensitive data protection

Assessment of direct and indirect sensitive-data pathways

Evidence
Context7 tool designTools never read local files or code themselves, limiting direct exposure; however, ContextCrush showed retrieved content can induce other tools to exfiltrate secrets, so isolation depends on the surrounding agent
mediumVerified: 2026-06-10
data minimization

Review of request payloads and tool surface area

Evidence
Context7 tool schemasOnly two narrowly-scoped read-only tools; requests carry just a library ID, topic, and token budget — a notably minimal data footprint among MCP servers
highVerified: 2026-06-10
third party data sharing

Data sharing pathway analysis across Upstash and LLM provider

Evidence
Upstash privacy policyQueries are processed by Upstash per its privacy policy; retrieved docs additionally flow to the connected LLM provider like all MCP tool results
mediumVerified: 2026-06-10
👁️Trust & Transparency
+
documentation quality

Documentation completeness and accuracy review

Evidence
Context7 READMEClear installation instructions for 20+ MCP clients, both local and remote transports, and well-documented tool parameters
highVerified: 2026-06-10
open source transparency

Source availability review of client/server versus hosted backend

Evidence
GitHub repositoryMCP server is MIT-licensed and open source; the backend indexing/retrieval pipeline at context7.com is proprietary, leaving part of the stack unauditable
highVerified: 2026-06-10
incident disclosure

Review of vendor communication during and after the ContextCrush incident

Evidence
Noma Labs — ContextCrush disclosureUpstash cooperated with coordinated disclosure and patched within 5 days; public acknowledgment came primarily through the researcher's publication rather than a detailed vendor advisory
highVerified: 2026-06-10
operation visibility

Logging and traceability assessment of tool calls and returned content

Evidence
Context7 tool designBoth tools are read-only with visible parameters and outputs in MCP host logs; retrieved doc content is fully inspectable before the agent acts on it
highVerified: 2026-06-10
⚙️Operational Excellence
+
ease of setup

Setup complexity assessment across remote and local installation paths

Evidence
Context7 installation docsRemote endpoint requires only a URL (https://mcp.context7.com/mcp) with no install; local server via npx @upstash/context7-mcp; no API key needed for basic use
highVerified: 2026-06-10
performance

Latency measurement of resolve and retrieval calls

Evidence
Context7 hosted APIDoc retrievals typically return in under a couple of seconds; configurable token budget keeps context costs predictable
mediumVerified: 2026-06-10
feature coverage

Feature scope assessment relative to documentation-retrieval needs

Evidence
Context7 tool referenceDeliberately narrow: two tools doing one job well (documentation retrieval); no code search, examples execution, or private-docs indexing in the open tier
highVerified: 2026-06-10
community adoption

Adoption metrics and ecosystem-integration analysis

Evidence
GitHub API57,127 stars as of 2026-06-10 — the largest MCP server repository on GitHub; integrated into setup guides of most major MCP clients
highVerified: 2026-06-10
maintenance activity

Commit frequency, release cadence, and patch-responsiveness analysis

Evidence
GitHub repository activityActive maintenance by Upstash with regular releases, fast security patching (ContextCrush fixed in 5 days), and continuous library-index growth
highVerified: 2026-06-10
Strengths
  • +Directly addresses hallucinated/outdated APIs with version-specific, current documentation
  • +Largest MCP server community on GitHub (57.1k stars) with broad client integration
  • +Minimal tool surface: two read-only tools with a small, predictable data footprint
  • +Zero-friction setup — remote endpoint works with just a URL, no API key required
  • +Fast vendor response to the ContextCrush vulnerability (patched in 5 days)
  • +Configurable token budget keeps documentation injection cost-controlled
Limitations
  • !ContextCrush (Feb 2026) proved the library index is an exploitable injection channel into agent contexts; retrieved third-party content remains untrusted by design
  • !Open publishing model means doc quality and integrity vary across the index
  • !Backend indexing/retrieval pipeline is proprietary and unauditable
  • !Dependent on Upstash's hosted service — no fully offline operation
  • !Queries reveal a team's technology stack to a third party
  • !Narrow scope: documentation retrieval only, no private-docs support in the open tier
Metadata
license: MIT
supported platforms
0: All platforms with Node.js 18+ (local server); any MCP client (remote endpoint)
programming languages
0: TypeScript
mcp version: 1.0
github repo: https://github.com/upstash/context7
github stars: 57127
package: @upstash/context7-mcp
remote endpoint: https://mcp.context7.com/mcp
api dependency: Context7 hosted documentation API (Upstash)
authentication: Optional API key (higher rate limits)
first release: 2025-04
maintained by: Upstash
status: Active
security incidents
0: [object Object]
transport types
0: stdio
1: streamable-http
installation methods
0: npm
1: npx
2: remote-url
3: docker

Use Case Ratings

code generation

Core use case — current, version-specific docs measurably reduce hallucinated APIs and deprecated patterns

research assistant

Excellent for researching library capabilities and APIs; limited to indexed open-source documentation

education

Strong for learning frameworks with accurate, up-to-date examples instead of stale training data

content creation

Useful for writing accurate technical tutorials and documentation-backed articles

data analysis

Indirectly helpful (correct API usage for analysis libraries) but not an analysis tool itself