Coding agents are already reading your docs.
Claim your library on Context7 to:
→ control what they read
→ see which agents and MCP clients use it
→ find doc gaps from real developer questions
→ get the verified badge
Free for every maintainer.
Claim your library on Context7 to:
→ control what they read
→ see which agents and MCP clients use it
→ find doc gaps from real developer questions
→ get the verified badge
Free for every maintainer.
1
3
16
Docs7 Example Check's are live.
It runs every TS, JS and Python example in your docs against the latest version of your code:
- broken examples get fixed and run again.
- all fixes land in one PR for you to review.
Try it below for free 👇
It runs every TS, JS and Python example in your docs against the latest version of your code:
- broken examples get fixed and run again.
- all fixes land in one PR for you to review.
Try it below for free 👇
2
3
22
We ran Exa's own coding search benchmark (WebCode) on Context7 Search.
Same accuracy as Exa 3.0: 85.7% vs 86.2%
~1/5 the tokens: 1,050 vs 5,000
~1/3 the wrong answers: 2.3% vs 7.3%
upstash.com/blog/context7-…
Same accuracy as Exa 3.0: 85.7% vs 86.2%
~1/5 the tokens: 1,050 vs 5,000
~1/3 the wrong answers: 2.3% vs 7.3%
upstash.com/blog/context7-…
2
4
23
Context7 Search: Exa for code.
Ask a question in plain English, get the right snippets from official docs. One GET request.
Drop it into your agent as a tool. Free plan = 1,000 searches/month.
Try it, no key needed:
Ask a question in plain English, get the right snippets from official docs. One GET request.
Drop it into your agent as a tool. Free plan = 1,000 searches/month.
Try it, no key needed:
1
3
32
Versioned docs now work for websites, not just repos.
Context7 detects versions on docs sites automatically and indexes each one. Your agent can query the exact version you’re on via MCP.
Context7 detects versions on docs sites automatically and indexes each one. Your agent can query the exact version you’re on via MCP.
4
3
12
Let agents search your docs with a single /ask endpoint
- No API key needed
- Returns the exact answers from docs
- Agents use it by themselves
- Built-in o11y
Agents find the answer in a single call instead of fetching page after page to answer the same question
- No API key needed
- Returns the exact answers from docs
- Agents use it by themselves
- Built-in o11y
Agents find the answer in a single call instead of fetching page after page to answer the same question
2
4
22
Docs7 PR Review agent can now commit fixes, not just comment.
Here it catches a broken link and a typo in a docs PR, pushes the fix to the branch, and the PR merges.
Here it catches a broken link and a typo in a docs PR, pushes the fix to the branch, and the PR merges.
1
4
24
Reactive Resume (44K ⭐) uses Docs7 agents.
A real dev asked Context7 how cover letters become PDFs. The docs answered badly.
Context7 Agent opened a PR to fix the docs. @KingOKings merged it.
A real dev asked Context7 how cover letters become PDFs. The docs answered badly.
Context7 Agent opened a PR to fix the docs. @KingOKings merged it.
1
2
9
Introducing Docs7 Agents.
They keep your docs fresh and open PRs for you to review:
- Code Watch: a PR merges in your code, it updates the docs
- PR Review: checks links, grammar, build health in every docs PR
- Context7 Agent: turns reader questions into docs fixes
- Example de examples still work with the latest version
Available on the free tier.
context7.com/docs/docs7/age…
They keep your docs fresh and open PRs for you to review:
- Code Watch: a PR merges in your code, it updates the docs
- PR Review: checks links, grammar, build health in every docs PR
- Context7 Agent: turns reader questions into docs fixes
- Example de examples still work with the latest version
Available on the free tier.
context7.com/docs/docs7/age…
4
12
65
Today in Context7:
- Smarter routing: "Haskell base" → Haskell, not a crypto project
- Unindexed version? Nearest stable docs, not an error
- Responses now start with the most relevant snippets, so agents hit the answer early
WebCode bench: 44% → 78% → 86% in two weeks
- Smarter routing: "Haskell base" → Haskell, not a crypto project
- Unindexed version? Nearest stable docs, not an error
- Responses now start with the most relevant snippets, so agents hit the answer early
WebCode bench: 44% → 78% → 86% in two weeks
1
2
9
Introducing Context7 Ask! 🎉
You can now see why AI agents visit your website
◆ AI agents ask your site a question instead of crawling
◆ See all questions agents ask about your product
◆ See which questions your website doesn't answer
You can now see why AI agents visit your website
◆ AI agents ask your site a question instead of crawling
◆ See all questions agents ask about your product
◆ See which questions your website doesn't answer
8
3
57
New in Context7: query-docs takes a language hint.
Your agent passes the language, and matching code examples rank higher.
No setup needed, agents pick it up automatically.
Top snippet in the right language: 24% → 83% on our benchmark.
Your agent passes the language, and matching code examples rank higher.
No setup needed, agents pick it up automatically.
Top snippet in the right language: 24% → 83% on our benchmark.
7
6
22
PR previews are now free on Docs7 🎉
Open a PR → Docs7 builds it → you get a live preview
This used to be a Pro feature, but thanks to our incredibly cost efficient @Cloudflare infra, we can give it to everyone.
Try it for free 👉 context7.com/docs7
Open a PR → Docs7 builds it → you get a live preview
This used to be a Pro feature, but thanks to our incredibly cost efficient @Cloudflare infra, we can give it to everyone.
Try it for free 👉 context7.com/docs7
1
3
33
We tested Context7 vs Exa on 44 API changes from 2026.
Only current docs:
Context7: 59%
Exa: 18-25%
Tokens per search:
Context7: 828
Exa: ~8,000
Exa is faster though. We build Context7, so questions, code and results are open.
upstash.com/blog/context7-…
Only current docs:
Context7: 59%
Exa: 18-25%
Tokens per search:
Context7: 828
Exa: ~8,000
Exa is faster though. We build Context7, so questions, code and results are open.
upstash.com/blog/context7-…
0
7
28
How we keep our docs agent friendly with Docs7 👇
- Every site has its own search endpoint
- All pages have a short note for agents to navigate faster
- Intuitive yet powerful agent-oriented analytics
Plus many small quality-of-life improvements
- Every site has its own search endpoint
- All pages have a short note for agents to navigate faster
- Intuitive yet powerful agent-oriented analytics
Plus many small quality-of-life improvements
1
5
23
Every site in Docs7 is way faster now 🚀
- 13x smaller size for median worker
- 5x faster startup times
- cold requests up to 3x faster
Live for all of our users, nothing to change on your side!
- 13x smaller size for median worker
- 5x faster startup times
- cold requests up to 3x faster
Live for all of our users, nothing to change on your side!
3
5
41
Context7 API overage drops to $5 / 1,000 calls (was $10), so more people can use Search API.
Context7 Search is Exa for code: it grounds your agent in up-to-date docs and code examples, so it stops guessing.
Pro includes 2,000 free calls.
New price starts October 1.
Context7 Search is Exa for code: it grounds your agent in up-to-date docs and code examples, so it stops guessing.
Pro includes 2,000 free calls.
New price starts October 1.
5
6
43
docs7 deploy
One command.
Your docs are hosted on *.docs7.io
Agents can read them through Context7.
One command.
Your docs are hosted on *.docs7.io
Agents can read them through Context7.
1
2
29
Introducing Context7 Search:
a grounding API for coding agents.
One GET request, get the docs snippets.
Unlike web search:
• Official docs, managed by library owners
• Scanned for prompt injection and malware
• Fast and token efficient
Blog:
upstash.com/blog/context7-…
a grounding API for coding agents.
One GET request, get the docs snippets.
Unlike web search:
• Official docs, managed by library owners
• Scanned for prompt injection and malware
• Fast and token efficient
Blog:
upstash.com/blog/context7-…
13
18
137
We benchmarked Docs7 vs Mintlify on Upstash docs.
Docs7 won 33 of 34 Pagespeed comparisons.
Page load on mobile: 3.3s vs 9.7s
Static HTML at build time, served from Cloudflare. Raw data in the post.
upstash.com/blog/docs7-vs-…
Docs7 won 33 of 34 Pagespeed comparisons.
Page load on mobile: 3.3s vs 9.7s
Static HTML at build time, served from Cloudflare. Raw data in the post.
upstash.com/blog/docs7-vs-…
2
2
32
We tested Jev against the models we use inside Context7's parsing pipeline (Gemini Flash, DeepSeek).
5 classification tasks. Results:
- 3 ties: query relevance, duplicate detection, website suitability
- 1 win: page classification — 85% vs 56%
- 1 loss: crawl-root selection — ster, 3-20x cheaper
Tagging a single page: Jev wins.
Reasoning about a whole site's structure: it doesn't.
Crawl-root selection: given a URL + site nav, pick which section to crawl.
Query relevance: is the user's question about this library or something else.
Duplicate detection: are two snippets the same example.
Website suitability: is this site technical docs worth indexing.
Page classification: does this doc page have code, API ref, or info.
5 classification tasks. Results:
- 3 ties: query relevance, duplicate detection, website suitability
- 1 win: page classification — 85% vs 56%
- 1 loss: crawl-root selection — ster, 3-20x cheaper
Tagging a single page: Jev wins.
Reasoning about a whole site's structure: it doesn't.
Crawl-root selection: given a URL + site nav, pick which section to crawl.
Query relevance: is the user's question about this library or something else.
Duplicate detection: are two snippets the same example.
Website suitability: is this site technical docs worth indexing.
Page classification: does this doc page have code, API ref, or info.
10
11
121
Context7 now indexes DeepWiki pages too.
Before: "the lifecycle of a user message in Chainlit" → nothing useful, because no docs page says it.
After: real answer, pulled from the repo's DeepWiki.
Your agent can now ask how things work.
Before: "the lifecycle of a user message in Chainlit" → nothing useful, because no docs page says it.
After: real answer, pulled from the repo's DeepWiki.
Your agent can now ask how things work.
1
3
17
Every Context7 doc is scanned before your agent sees it
Your AI agent reads docs. Docs can carry prompt injections.
Context7 scans every doc before indexing. A custom classifier flags injection attempts and malware patterns.
context7.com/docs/security/…
Your AI agent reads docs. Docs can carry prompt injections.
Context7 scans every doc before indexing. A custom classifier flags injection attempts and malware patterns.
context7.com/docs/security/…
0
3
8
Every Docs7 site serves WebMCP out of the box. No setup.
And the dashboard shows what AI agents are actually searching for in your docs.
And the dashboard shows what AI agents are actually searching for in your docs.
3
6
32
Doc7 shows you which AI agents read your docs.
Every request from ChatGPT, Claude, Meta, etc. is a row: agent, page, country, time.
Screenshot is from Upstash docs, right now. Turns out agents read a lot of ratelimit and QStash pages.
Every request from ChatGPT, Claude, Meta, etc. is a row: agent, page, country, time.
Screenshot is from Upstash docs, right now. Turns out agents read a lot of ratelimit and QStash pages.
0
5
17
Context7 is now an OpenCode plugin.
opencode plugin @upstash/context7-opencode
MCP server + a skill that auto-triggers doc lookups. One command, no config.
context7.com/docs/clients/o…
opencode plugin @upstash/context7-opencode
MCP server + a skill that auto-triggers doc lookups. One command, no config.
context7.com/docs/clients/o…
5
20
256
Introducing Docs7 - the best way to serve documentation to agents.
◆ Automatically serve clean, agent-readable docs
◆ Native WebMCP and skills support
◆ Deep insights into AI traffic & searches
We make agents love your product 👇
◆ Automatically serve clean, agent-readable docs
◆ Native WebMCP and skills support
◆ Deep insights into AI traffic & searches
We make agents love your product 👇
23
67
698
Context7 without MCP:
GET /api/v2/libs/search
GET /api/v2/context
paste into your prompt
No install. No client. Just HTTP.
GET /api/v2/libs/search
GET /api/v2/context
paste into your prompt
No install. No client. Just HTTP.
6
33
349
New blog: Context7 vs Claude Code Web Search 🎉
◆ Context7 uses ~99% less input tokens
◆ also 37% fewer tokens and 35% lower cost
The easiest way to make your agent write better code 👇
◆ Context7 uses ~99% less input tokens
◆ also 37% fewer tokens and 35% lower cost
The easiest way to make your agent write better code 👇
6
3
30
Introducing Context7 Suggestions! 🎉
See what's missing in your docs based on real user questions
◆ Lists user questions your docs couldn't answer
◆ AI researches the right answers
◆ Shows what to add to your docs
See what's missing in your docs based on real user questions
◆ Lists user questions your docs couldn't answer
◆ AI researches the right answers
◆ Shows what to add to your docs
3
4
68
Introducing Context7 CLI! 🎉
The modern, token-friendly way to get up-to-date documentation for AI.
90-second demo 👇
The modern, token-friendly way to get up-to-date documentation for AI.
90-second demo 👇
25
54
735
1000+ companies and devs — including Amazon and Google — have claimed their libraries on Context7.
Claim yours to:
• See agent usage analytics
• Manage versions
• Refresh your library anytime
Just add context7.json to your docs.
Claim yours to:
• See agent usage analytics
• Manage versions
• Refresh your library anytime
Just add context7.json to your docs.
1
6
41
Context7:
• Does not ingest user code
• Does not store conversation history
• Uses only minimum data for retrieval
Runs on SOC 2–compliant infrastructure.
Enterprise supports SSO and audit trails.
• Does not ingest user code
• Does not store conversation history
• Uses only minimum data for retrieval
Runs on SOC 2–compliant infrastructure.
Enterprise supports SSO and audit trails.
0
0
0
Anyone can submit documentation.
Context7 runs a two-pass prompt-injection detection pipeline:
• Blocks malicious submissions
• Reduces false positives
• Allows benign but suspicious-looking snippets
Context7 runs a two-pass prompt-injection detection pipeline:
• Blocks malicious submissions
• Reduces false positives
• Allows benign but suspicious-looking snippets
1
0
0
Libraries and skills can be verified.
Verification criteria:
- Trust score ≥ 9
- Top-100 by MCP usage with trust ≥ 6
- Claimed by the library owner
Currently visual. Planned to affect ranking directly.
Verification criteria:
- Trust score ≥ 9
- Top-100 by MCP usage with trust ≥ 6
- Claimed by the library owner
Currently visual. Planned to affect ranking directly.
1
0
0
Automatic ranking is not the only control.
Teams can explicitly restrict retrievable libraries via Public Library Access in the dashboard.
Useful for org-wide policy enforcement.
Teams can explicitly restrict retrievable libraries via Public Library Access in the dashboard.
Useful for org-wide policy enforcement.
1
0
0
Context7 assigns trust scores based:
• Org size, stars, account age
• Activity level, followers, profile completeness
• Org size, stars, account age
• Activity level, followers, profile completeness
1
0
0
Documentation often contains multiple versions.
Context7:
• Detects versioned content
• Excludes older versions
• Indexes only the most recent one
Prevents stale or conflicting references.
Context7:
• Detects versioned content
• Excludes older versions
• Indexes only the most recent one
Prevents stale or conflicting references.
1
0
0
Retrieval includes deduplication.
Pipeline:
• Exact match detection
• Cosine similarity for near-duplicates
• Final filter to confirm true redundancy
Removes repeated snippets and limits context inflation.
Pipeline:
• Exact match detection
• Cosine similarity for near-duplicates
• Final filter to confirm true redundancy
Removes repeated snippets and limits context inflation.
1
0
0
Context7 benchmarks libraries.
• Generate library-specific developer questions
• Measure how well docs answer them
• Use scores to prioritize retrieval
• Expose weak coverage to maintainers
Benchmarks measure answerability, not popularity.
• Generate library-specific developer questions
• Measure how well docs answer them
• Use scores to prioritize retrieval
• Expose weak coverage to maintainers
Benchmarks measure answerability, not popularity.
1
0
0
Quality & safety mechanisms in Context7.
Benchmarks, trust scores, deduplication, version analysis, verification, safety.
Details in replies ↓
Benchmarks, trust scores, deduplication, version analysis, verification, safety.
Details in replies ↓
1
2
12
Context7 Quality (3/9): Version Analyzer
• Detects multiple versions in documentation
• Filters out older or deprecated versions
• Keeps only the latest docs for retrieval
• Avoids outdated or duplicate context
• Detects multiple versions in documentation
• Filters out older or deprecated versions
• Keeps only the latest docs for retrieval
• Avoids outdated or duplicate context
1
2
13
Context7 Quality (2/9): Deduplication
• Retrieval is not only benchmarks
• We remove duplicate and near-duplicate snippets
• Exact match → similarity check → final filter
• Less noise, more useful context
• Retrieval is not only benchmarks
• We remove duplicate and near-duplicate snippets
• Exact match → similarity check → final filter
• Less noise, more useful context
2
3
31
Context7 Quality (1/9): Benchmark Scores
• Libraries are evaluated using generated benchmark questions
• Library owners can add and update these questions
• Each library receives a benchmark score
• Scores directly affect retrieval priority
• Libraries are evaluated using generated benchmark questions
• Library owners can add and update these questions
• Each library receives a benchmark score
• Scores directly affect retrieval priority
3
2
13
NEW: Create custom agent skills
◆ Choose what your agent should be good at
◆ Context7 reads the docs and creates the skill
◆ Auto-installs into your project
Quick demo 👇
◆ Choose what your agent should be good at
◆ Context7 reads the docs and creates the skill
◆ Auto-installs into your project
Quick demo 👇
14
53
466
Introducing Context7 Skills! 🎉
◆ We extracted 24k skills from 65k repos
◆ Skills for Tailwind, React, Better-Auth, etc.
◆ Install in a single CLI command
Perfect for Cursor, Claude Code & others 👇
◆ We extracted 24k skills from 65k repos
◆ Skills for Tailwind, React, Better-Auth, etc.
◆ Install in a single CLI command
Perfect for Cursor, Claude Code & others 👇
59
186
2,192
New blog: How we built OAuth for our MCP server 🎉
◆ A complete OAuth + MCP guide
◆ Probably the most in-depth guide you'll find
◆ With diagrams and code examples
Step-by-step, link below 👇
◆ A complete OAuth + MCP guide
◆ Probably the most in-depth guide you'll find
◆ With diagrams and code examples
Step-by-step, link below 👇
1
8
19
Prompt injection is a real attack surface for LLM-based systems.
Context7 treats docs as untrusted input: every snippet is scanned and quarantined on injection detection.
Context7 treats docs as untrusted input: every snippet is scanned and quarantined on injection detection.
1
10
84
We fixed Context7 biggest problem: Context bloat.
Introducing the new Context7 🎉
◆ 65% less tokens per query
◆ 38% faster queries
◆ Better output quality
Introducing the new Context7 🎉
◆ 65% less tokens per query
◆ 38% faster queries
◆ Better output quality
31
75
1,043
Context7 pulls documentation from thousands of sources. Some try to inject malicious code.
Here's how we prevent that:
◆ Auto-detect code injections
◆ Score & benchmark every library's docs
Full blog article w/ examples below 👇
Here's how we prevent that:
◆ Auto-detect code injections
◆ Score & benchmark every library's docs
Full blog article w/ examples below 👇
3
4
57