RepospecMCP server for Claude Code & Cursor
The spec exists.
Your AI just can't see it.
Repospec gives Claude Code and Cursor one-call fetch and search over the specs in your private repos. Built for developers with too many repositories to remember where anything is written — the "where did I write that?" moment that hits several times a day.
You're on the list — we'll email you at launch. Please make sure you can receive emails from repospec.dev.
Something went wrong. Please try again in a moment.
Pre-launch. No spam — one email when we open the doors.
The multi-repo dilemma
Index everything, or read everything?
You're working in Service A. The spec you need lives in Service B's repo. It's written down — your AI just can't reach it. Until now, you've had two options, and both charge a tax.
Add the repo to your workspace
Multi-root workspaces work — but every folder you open becomes part of the index. Irrelevant files leak into context, searches get noisy, and edits land where you didn't ask.
cost: noise
Fetch files on demand
Fine when you remember the exact path. But "I know I wrote this somewhere" means the AI reads through candidate files to find it — loading full contents just to run a search.
cost: tokens
The third way: move only the search outside
Repospec keeps the full text of your registered specs cached and searched in a single DB query, and returns just the matching lines plus a few lines of context. One tool call, one snippet — no GitHub Code Search rate-limit roulette, no full-file dumps into your context window. Your repos stay the source of truth; Repospec is read-only plumbing.
Setup in minutes
Three steps, then forget it exists
Connect GitHub
Install the GitHub App and pick the repos where your specs live. Read-only access — no personal access tokens to create or babysit.
Register your spec files
Tell Repospec which files are specs — docs/api.md, spec/openapi.yaml, whatever your convention is. Only these are indexed. Nothing else is touched.
Add one line to your MCP config
Issue a token and paste the snippet. Done.
What your AI gets
Three tools. That's the whole surface.
Your assistant calls them on its own whenever it needs a spec — you don't have to think about it.
list_spec_filesLists every registered project and its spec files, so the AI knows what exists before it asks.
args: none
read_specFetches the spec files for a project — one file or all of them — straight from its GitHub repository, cached for speed.
args: project, path?
search_specsCross-repo keyword search over every spec you've registered. Returns matching lines with surrounding context, not whole files.
args: query, project?
You can do most of this for free.
Here's what's left.
The remote GitHub MCP Server is genuinely good: OAuth once, fetch files from private repos, even search them. If you know your file paths, it covers you — for free, and we'll say so. What Repospec sells isn't a feature GitHub lacks. It's the removal of friction you'd otherwise hit several times a day.
| Daily friction | GitHub MCP Server | Repospec |
|---|---|---|
| Search rate limit | Code search API: 10 requests/min — an AI's trial-and-error loop drains it fast | Cached search — no GitHub Code Search quota. 100/min per token (planned) |
| Round trips | Minimum two: search, then fetch each result to read it | One call returns the snippet itself |
| Search scope | Default branch only, files under 384 KB only | Registered spec files, cached for cross-repo search |
| Context noise | The whole repo is in scope — irrelevant files leak into results | Only files you marked as specs; only matching lines + context are returned |
| Team rollout | Each member builds and maintains their own rules + MCP setup | Share a token, manage from one dashboard (Team plan, coming later) |
GitHub Code Search limits (10 req/min, default branch, 384 KB) and GitBook pricing ($65+/mo) are from the vendors' docs as of July 2026; they may change. Sources: GitHub Search API, GitBook pricing.
None of these are "it doesn't work." They're all "it never feels quite right — every single day." "Where did I write that?" happens several times a day; $8/month is what it costs to stop stepping on it. (And if you want a full docs platform — editing, publishing, versioning — that's GitBook at $65+/month. Repospec deliberately isn't that.)
Built for private repos
Security you can explain to your team
No certifications to hide behind — so the mechanism is fully documented instead.
Read-only GitHub App
The App requests Contents: Read-only and webhooks — nothing else. It cannot write to your repos, and you never hand over a PAT.
Hashed tokens, shown once
MCP tokens are stored as SHA-256 hashes. The raw value is displayed exactly once at creation, then masked forever.
Revoke anytime, audit at a glance
Every token shows its last-used time, and any token can be revoked instantly from the dashboard.
Cache that deletes itself
Spec content is cached only for speed and search. Delete a repo or your account, and the cache is removed with it — automatically, at the database level. All traffic is TLS end to end.
Pricing
One flat price. No seats, no surprises.
You're not buying features the free tools lack — you're buying the removal of friction you hit daily. About a quarter a day.
PersonalAt launch
$8 / month
- Unlimited projects
- 200 spec files
- 5 MCP tokens
- Everything in Free
TeamComing soon
— / month
- Coming soon
Downgrading never deletes your data. Anything over the new plan's limits is paused — not removed — and comes back the moment you're within limits again.
FAQ
Honest answers
Do I actually need this?
Maybe not — and we'd rather tell you now. If your project is small enough that you remember every file path, write your spec paths into .cursor/rules and let the remote GitHub MCP Server read them: one OAuth, no Docker, no PAT, and it's free. Repospec starts to earn its keep when "multiple repos" × "I can't remember where I wrote what" becomes your daily reality.
Can't GitHub MCP Server do this for free?
Almost, yes. It fetches files from private repos and can even search them. The difference is friction — all of it documented GitHub behavior: code search is capped at 10 requests/minute (a separate, much tighter bucket than the regular 5,000/hour), it only covers the default branch and files under 384 KB, and results are metadata — reading the content takes a second round trip. None of that blocks you. All of it grates when you hit it several times a day. With Repospec, search is one call against the cache, one snippet — no GitHub Code Search quota. A 100 requests/minute per token limit is planned. Sources for these limits are listed under the comparison table above.
Does Repospec store my code?
Not your code — only the spec files you register are cached, for fast fetching and search. Your repos remain the single source of truth; Repospec has read-only access and no editing features by design.
Which AI tools are supported?
Anything that speaks MCP over Streamable HTTP — Claude Code and Cursor are the primary targets and get first-class setup snippets.
What happens when a spec changes?
When you push to your repo, GitHub notifies Repospec via webhook and the cached copy of the changed spec files is updated automatically. After that refresh, your AI sees the new content on the next tool call — both fetch and search.
Stop being your AI's courier.
Be first in line when Repospec launches.
One email, no drip campaign.
You're on the list — we'll email you at launch. Please make sure you can receive emails from repospec.dev.
Something went wrong. Please try again in a moment.